w%I

TECHNOLOGY:

UNIVAC 1108EXEC LEVEL 32R2PERFORMANCE HANDBOOK

00-34

NBS Special Publication 500-34

U.S. DEPARTMENT OF COMMERCENational Bureau of Standards

NATIONAL BUREAU OF STANDARDS

The National Bureau of Standards' was established by an act of Congress March 3, 1901. TheBureau's overall goal is to strengthen and advance the Nation's science and technology and

facilitate their effective application for public benefit. To this end, the Bureau conducts

research and provides: (1) a basis for the Nation's physical measurement system, (2) scientific

and technological services for industry and government, (3) a technical basis for equity in

trade, and (4) technical services to promote public safety. The Bureau's technical work is

performed by the National Measurement Laboratory, the National Engineering Laboratory,

and the Institute for Computer Sciences and Technology.

THE NATIONAL MEASUREMENT LABORATORY provides the national system of

physical and chemical and materials measurement; coordinates the system with measurement

systems of other nations and furnishes essential services leading to accurate and uniform

physical and chemical measurement throughout the Nation's scientific community, industry,

and commerce; conducts materials research leading to improved methods of measurement,

standards, and data on the properties of materials needed by industry, commerce, educational

institutions, p'^.- '^"vernment; provides advisory and research services to other Government

Agencies; develops, produces, and distributes Standard Reference Materials; and provides

calibration services. The Laboratory consists of the following centers:

Absolute Physical Quantities^ — Radiation Research — Thermodynamics and

Molecular Science — Analytical Chemistry — Materials Science.

THE NATIONAL ENGINEERING LABORATORY provides technology and technical

services to users in the public and private sectors to address national needs and to solve

national problems in the public interest; conducts research in engineering and applied science

in support of objectives in these efforts; builds and maintains competence in the necessary

disciplines required to carry out this research and technical service; develops engineering data

and measurement capabilities; provides engineering measurement traceability services;

develops test methods and proposes engineering standards and code changes; develops and

proposes new engineering practices; and develops and improves mechanisms to transfer

results of its research to the utlimate user. The Laboratory consists of the following centers:

Applied Mathematics — Electronics and Electrical Engineering^ — Mechanical

Engineering and Process Technology^ — Building Technology — Fire Research —Consumer Product Technology — Field Methods.

THE INSTITUTE FOR COMPUTER SCIENCES AND TECHNOLOGY conducts

research and provides scientific and technical services to aid Federal Agencies in the selection,

acquisition, application, and use of computer technology to improve effectiveness and

economy in Government operations in accordance with Public Law 89-306 (40 U.S.C. 759),

releva'nt Executive Orders, and other directives; carries out this mission by managing the

Federal Information Processing Standards Program, developing Federal ADP standards

guidelines, and managing Federal participation in ADP voluntary standardization activities;

provides scien^tific and technological advisory services and assistance to Federal Agencies; and

provides the technical foundation for computer-related policies of the Federal Government.

The Institute consists of the following divisions:

Systems and Software — Computer Systems Engineering — Information Technology.

'Headquarters and Laboratories at Gaithersburg, Maryland, unless otherwise noted;

mailing address Washington,D.C. 20234.

'Some divisions "within the center are located at Boulder, Colorado, 80303.

The National Bureau of Standards was reorganized, effective April 9, 1978.

COMPUTER SCIENCE & TECHNOLOGY:

Performance Handbook

iSATIONAL BUREAU.OF STANDARDS

JUL 1 3 1978

f]p-+ <X4L(L.

UNIVAC 1108 EXEC LEVEL 32R2 us?

Capt. John C. Kelly and Lt. Gary P. Route ^(^CQ^\^ \

Federal Computer Performance Evaluation and

Simulation Center

Washington, DC 20330

Sponsored by

U.S. Military Personnel Center

Alexandria, VA 22332

U.S. DEPARTMENT OF COMMERCE, Juanita M. Kreps, Secretary

Dr. Sidney Harman, Under Secretary

Jordan J. Baruch, Assistant Secretary for Science and Tecinnology

VyV ^ NATIONAL BUREAU OF STANDARDS, Ernest Ambler, Director

Issued June 1978

Reports on Computer Science and Technology

The National Bureau of Standards has a special responsibility within the Federal

Government for computer science and technology activities. The programs of the

NBS Institute for Computer Sciences and Technology are designed to provide ADPstandards, guidelines, and technical advisory services to improve the effectiveness of

computer utilization in the Federal sector, and to perform appropriate research and

development efforts as foundation for such activities and programs. This publication

series will report these NBS efforts to the Federal computer community as well as to

interested specialists in the academic and private sectors. Those wishing to receive

notices of publications in this series should complete and return the form at the end

of this publication.

National Bureau of Standards Special Publication 500-34

Nat. Bur. Stand. (U.S.) Spec. Pub!. 500-34, 144 pages (June 1978)

CODEN: XNBSAV

Library of Congress Catalog Card Number: 78-600054

U.S. GOVERNMENT PRINTING OFFICE

WASHINGTON: 1978

For sale by the Superintendent of Documents, U.S. Government Printing Office, Washington, D.C. 20402

Stock No. 003-003-01948-8 Price 3.25

(Add 25 percent additional for other than U.S. mailing).

FOREWORD

Operating systems largely determine the performance ofmodern computer systems „ They control large-scale, multi-programming computers with programmed rules and proceduresthat may be tailored, within limits, to fit the particularneeds of different users in widely differing applications.Proper tuning of operating system control variables is whereany serious attempt to achieve efficient computer utilizationmust begin, and often where its greatest benefits accrue.

This report describes the user-adjustable controls ofone such operating system: UNIVAC EXEC 8, level 32 release2. It presents methods of detecting and diagnosing perfor-mance problems, identifies important EXEC control points,and recommends control settings that are believed to producegenerally satisfactory results at some installations.

Such information is, of course, site and system specific.That is, it has direct application only to EXEC 8 sites,limited application outside of level 32, and qualifiedapplication even to other sites with the identical release.For non-EXEC sites, the document should prove of interestprimarily as a format for presenting a kind of informationthat would be useful in any operating system environmentand as a model for analyzing the subtle problem of operatingsystem tuning on any vendor system. For other releases ofEXEC 8, resemblances to level 32 are probably sufficient tostimulate creative thinking about specific areas of perfor-mance improvement, although the caution against careless oruncritical application of recommendations is especiallyurgent here. For sites using level 32, the informationhere should be of enormous value and much of it directlytransferable. This class of reader must remember only thatoperating systems are adjustable precisely because no twoinstallations have quite the same configuration, workloador functional priorities, and that no set of controlparameters will suit all implementations even of the samesystem equally well.

This report was prepared in its original form by theFederal Computer Performance Evaluation and SimulationCenter (FEDSIM) under the sponsorship of the U.S. ArmyMilitary Personnel Center (MILPERCEN) . It is reprintedhere with minor editorial changes by the National Bureau ofStandards' Institute for Computer Sciences and Technology(NBS/ICST) in the belief that it will have useful applicationat other installations besides the one for which it wasintended.

iii

Publication of this report in no way implies recommen-dation or endorsement of UNIVAC EXEC 8 systems by theNational Bureau of Standards, nor is it to be construed topass qualitative judgment in any degree on the performancecharacteristics of the system or any system features describeherein

.

Special note is made of the fact that this publicationis the result of a cooperative interaction between NBS andFEDSIM to make the results of FEDSIM's generally applicablework available to the Federal ADP community. NBS alsowishes especially to express its appreciation to MILPERCENfor making the occasion for this collaboration possible.

The authors of this report welcome questions concerningthe technical content, which should be addressed to them at:FEDSIM, Washington, DC 20330. NBS/ICST would appreciatevery much comments concerning the general utility and needfor computer system tuning guides. Of particular interestwould be proposed or candidate reports, facilities interestedin sponsoring or participating in developing tuning guides,and facilities interested in serving as test beds forevaluating candidate guidelines. Queries and expressionsof interests of this nature should be addressed to eitherJohn F. Wood or Richard Fo Dunlavey, National Bureau ofStandards, Building 225, RoomA265, Washington, DC 20234(Area Code 301, 921-3485).

M. Zane Thornton, Acting DirectorInstitute for Computer Sciencesand Technology

National Bureau of StandardsU, S. Department of Commerce

iv

PREFACE

This report is based on a detailed analysis of theUNIVAC 1108 Operating System - EXEC Level 32R2 - as operatedby the United States Army Military Personnel Center(MILPERCEN) . Since the procedures and guidelines specifiedin the report strongly depend on the particular equipmenttype and EXEC level, care should be taken before applyingthe results to different environments. Questions relatingto the subject of this report or to the possibility of ex-tending the results should be addressed to the study'sauthors at the Federal Computer Performance Evaluation andSimulation Center (FSDSIM)

.

V

ABSTRACT

This is a report prepared for the Army Military-Personnel Center (MILPERCEN) . It describes a set ofhypotheses for evaluating the performance of UNIVAC 1100Computer Systems. The hypotheses specifically apply toEXEC Level 32R2 operating on UNIVAC 1108 Computer Systems.Attempts to apply the guidelines to different UNIVAC 1100models or to different levels of the EXEC may lead toerroneous results.

The report contains sections on each major complexwithin the EXEC. Each section (1) states the performancehypotheses associated with that complex, (2) explains howthe available measurement tools may be used to determine ifthe hypotheses are true, and (3) identifies which performanceparameters need to be changed to correct the problem situ-ation. All of the performance hypotheses discussed in thebody of the report are collected together in Appendix A.Appendix B summarizes the performance parameters andAppendix C contains an introduction to the Software Instru-mentation Package (SIP)

.

Key words: Computer performance management; measurementtools; operating system performance; performance guidelines;performance measurement; tuning guides; UNIVAC 1108.

VI

TABLE OF CONTENTS

Section 'Page

I. INTRODUCTION „ 1

A. Background 1

B . Ob j ect ive <, 1

II. METHODOLOGY „ „ 3

III. PROCESSOR MANAGEMENT. .. „ 5

A. Introduction. o o o 5

Bo CPU Priorities. » o „ 7

C . CPU Quanta. „ o 8

D. Test and Set Interrupts o. 9

IV. MEMORY MANAGEMENT . . „ „ o . . 11A. Memory Configuration 11B. Memory Allocationo 11C. Demand/Batch Sharing „.o 21D. Function/Segment Activity 23E. EXPOOL. „...,„ 25

V. FACILITIES MANAGEMENT 29A. Master File Directory 29B. Mass Storage Rollouts 31C. Granule Tables 37D. Mass Storage Management Practices 40E. Scheduling o 43

VI. INPUT/OUTPUT. .. o . o .. o O...0 47A. Late Acknowledgments o 47B. Seek Time. . .

.

. . . 52C . Block Size. „ „ . » 55

VII. TEST AND DEBUG , o. 59

APPENDICES

A. PERFORMANCE HYPOTHESES » 61B . PERFORMANCE PARAMETERS . . . „ 67C. INTRODUCTION TO SIP „.„ 10 7

vii

TABLE OF CONTENTS

TABLES

Tab le Page

III-l Activity Priority Types „ 7

III-2 CPU Quanta 8

IV- 1 Core Priorities o o. 14IV-2 Demand Priority Levels for CQFACT=800 15IV- 3 Demand Priority Levels for CQFACT=3200 16IV-4 Batch Priority Levels 17IV-5 Cost Table , » . . o . 21IV-6 Segment Status Values o 25V-1 Mass Storage Thresholds 31V-2 Point Values for TSLR and MTBR 36V-3 Point Values for SIZE„ „ 36

VII-1 Test and Debug Parameters o.o.... 59B- 1 Performance Parameters 70C-1 SIP Reports (Level 4 . 0) . . . » o 110

FIGURES

Figure

III-l Nesting of EXEC Functions 5

III-2 Major System Queues ... o 6

IV- 1 Memory Configuration o 12IV-2 Demand/Batch Sharing Algorithm 22

viii

I. INTRODUCTION

A. BACKGROUND

The United States Army Military Personnel Center(MILPERCEN) is responsible for maintaining the personnelrecords of all enlisted and officer personnel in the Army.In support of this mission, MILPERCEN operates seven UNIVAC1108 computer systems. During FEDSIM's study, the computersystems were operating under control of the UNIVAC 1100Operating System, Level 32R2.

Because MILPERCEN recognized the importance of operatingsystem performance on overall system performance, theyrequested that FEDSIM study how operating system performancecould be improved. Their request for assistance was formal-ized in FEDSIM/MILPERCEN Customer Agreement AY-505-080-ARJ^IY

,

titled MILPERCEN EXEC-8 ANALYSIS. Additional tasks insupport of this project were outlined in Amendment No. 1 toCustomer Agreement AY-505-080-ARMY . The funding for theadditional tasks was obtained by transferring funds fromanother project, NA-505-079-ARf4Y

.

B. OBJECTIVE

The objective of this project was to optimize the perfor-mance of the UNIVAC 1100 Operating System for MILPERCEN'

s

unique environment. To accomplish this objective, theproject was divided into two parts: (1) the development ofa set of general performance guidelines and (2) the applica-tion of those guidelines to MILPERCEN. This report summarizesthe guidelines developed. A second report describes theirapplication to MILPERCEN.

In support of the project objective, the customeragreement outlined 14 areas of investigation:

1. Library Configuration2. Swapping3. EXEC-8 Configuration Parameters4. Queue Management/Interrupt Processing5. Internal Priority Management6. Real-Time Interface Tailoring7. Dynamic Allocation of Core Memory8. Management of Buffer Space9. Dispatching Time Slice Management

10. Data File Rollout/Rollback Management11. Communications Terminal Network12. Officer Personnel Utilization System13. I/O Complex14. Remote Terminal Emulator

After preliminary analysis, it was found that many of theseareas overlapped with one another. To eliminate the redun-dancies, the following consolidated areas were developed:

(1) Officer Personnel Utilization System(2) Processor Management(3) Memory Management(4) Facilities Management(5) Input/Output(6) Test and Debug

The Officer Personnel Utilization System is not discussed inthis report. It is treated separately in another report.This report devotes a separate section to each of theremaining areas mentioned above.

2

II. METHODOLOGY

A four-step procedure was used to prepare this set ofguidelines

:

(1) Learn the fundamentals of how each EXEC complexworks

.

(2) Identify the controllable performance parametersassociated with each complex.

(3) Outline the performance hypotheses associatedwith each complex.

(4) Determine how SIP can be used to decide if thehypotheses are true.

This procedure was followed for each EXEC complex. Resultsfor each complex are presented in separate sections.

The most difficult problem associated with this projectwas obtaining adequate documentation. The primary source ofinformation was the actual EXEC program listings. Additionalinformation was obtained from the UNIVAC EXEC InternalsCourse, the UNIVAC Dump Analysis Course, and the Programmer'sReference Manual.

The hypotheses presented in this report describe condi-tions which, if true, would have an adverse affect on perfor-mance. They are stated as if they were true, though theirexistence must be proven or disproven by empirical analysis.The hypotheses are not intended to be statements of fact,but rather pointers to potential performance problems.

I

3

III. PROCESSOR MANAGEMENT

A. INTRODUCTION

Processor management is the responsibility of theDispatcher. The Dispatcher is the heart of the operatingsystem; it makes the final decision about which activity toexecute next. Before a run can be dispatched, it must beprocessed by other parts of the EXEC. The Course Scheduler(CS) decides which runs should be opened; the Dynamic Allo-cator (DA) decides which activities associated with thoseruns should be loaded into main storage; and the Dispatcher(DISP) decides which activity, of those loaded into mainstorage, should be given control of the CPU. The nestedrelationship between the EXEC functions is illustrated inFigure III-l. When a run enters the system, it must passthrough a series of queues before it eventually completesexecuting. A queue is associated with each major EXECcomplex (see Figure III-2). This section addresses only theDispatcher Queue, also known as the Switch List (SWL) . Theother queues are discussed in following sections.

The three main performance hypotheses associated withprocessor management and activity dispatching are listedbelow:

(1) Batch programs that are assigned the same CPUpriority as demand programs are causing demandprograms to have degraded response times.

SCHEDULING

DYNAMIC ALLOCATION

DISPATCHING

______NESTING OF EXEC FUNCTIONS

FIGURE III-l

5

6

(2) CPU quanta that are set either too high or toolow are causing the CPU to be inefficiently allocated.

(3) An excessive number of test and set interruptsare causing programs to be delayed until data areasare unlocked.

Each of these problems is discussed in the sections thatfollow.

B. CPU PRIORITIES

All activities in the system are controlled by an inter-nal priority referred to as type and level (TAL) . TAL isinterpreted by the Dispatcher as a CPU priority. Typespecifies the priority class of an activity, and levelspecifies a subpriority within a type. There are sevenactivity types numbered 0 through 6. The number of allow-able levels varies for each type. TAL is denoted by a 3-digitoctal number of the form TLL, where T is the type and LL isthe level.

All activities are queued on the switch list accordingto their TAL (CPU priority) . Whenever the state of thesystem changes, the Dispatcher selects the highest priority(lowest TAL) activity on the svzitch list to execute next.

Activity Types 0, 1, and 3 are reserved for EXEC activ-ities. Type 2 is reserved for real-time activities; allother user activities operate at Types 4, 5, and 6. Theactivities associated with each type are summarized inTable III-l. Type 4 is reserved for Transaction InterfacePackage (TIP) programs. Both demand and batch programs runat Type 6

.

TYPE LEVELS ACTIVITIES

0 00 Interlock Processing (EXEC-0)1 01-02 High EXEC (EXEC-1)2 01-35 Real-Time3 01-35 Low EXEC (EXEC- 3)

4 01-07 TIP5 01-07 Deadline Batch6 01-07 Demand and Batch

ACTIVITY PRIORITY TYPES

TABLE III-l

7

This priority structure can cause performance problemsfor demand programs. Since demand and batch programs havethe same CPU priority, demand programs may be preempted bybatch programs. In previous versions of the EXEC, demandprograms executed at Type 4; but, when TIP was introduced,they were reduced to Type 6. For installations that do notrun TIP, demand programs may be changed back to Type 4 bychanging the software parameter DSWTCHTYP. If DSWTCHTYP ischanged to 4, batch throughput should be closely monitoredsince demand programs may lock out batch programs.

C. CPU QUANTA

The length of time an activity is given access to asystem resource is referred to as a quantum. There are twotypes of quanta: core quanta and CPU quanta. The DynamicAllocator allocates main storage in terms of core quanta,and the Dispatcher allocates CPU time in terms of CPU quantaSince an activity must be core-resident before it can bedispatched, it can receive a CPU quantum only if it iscurrently allocated a core quantum. Core quanta and theirperformance implications are discussed in a subsequentchapter

.

Only Program Types 4, 5, and 6 are assigned CPU quanta;Types 0, 1, 2, and 3 are allowed to execute until completionor until they are preempted by a higher priority activity.Quantum assignments are made according to the current levelof the activity. The quanta assigned to each user level aresummarized in Table III-2. These values are hardcoded inthe element DISP.

USER LEVELQUANTUM

(MILLISECONDS)

1 1.42 3

3 324 645 1286 2567 512

CPU QUANTA FOR TYPES 4, 5, AND 6

TABLE III-2

8

An activity initially begins execution at Level 2 and is

allowed to execute for the corresponding CPU quantum time.Level 1 is reserved for the processing of I/O interruptsthat occur at the completion of an I/O operation (101$ andIOWI$) . If the activity does not complete execution beforeits quantum expires, the level of the activity is incre-mented; and the activity is requeued on the switch list.This continues until the activity either completes executionor reaches Level 7. If the activity reaches Level 7, itcontinues there until it completes execution. One exceptionto the above procedure occurs when an activity voluntarilygives up control to do an I/O operation. In this case, theactivity returns to Level 2, regardless of its currentlevel.

CPU Dispatching is purely preemptive. An executingactivity may be preempted by a higher priority (TAL) activityat any time. After each interrupt, the Dispatcher mustdetermine the highest priority activity and give control toit. If the preempted activity has not exceeded its quantum,it will be requeued at the same level. The next time it isgiven control, it will be allowed to execute fo'r the quantumtime it has remaining.

CPU quanta are particularly sensitive to an installation'sworkload. If they are set too low, the CPU will spend adisproportionate percent of its time switching betweenactivities. Because of the large number of times theDispatcher is executed, this can represent a substantialamount of EXEC overhead. The overhead will show up in a SIPreport as high EXEC 0 percentages. If the quanta are settoo high, compute bound programs may degrade the responsetime of interactive programs. A delicate balance can onlybe achieved by experimentation, but the process is simplifiedwith Level 33 of the EXEC. The SIP report for Level 3 3

prints the number of times an activity exceeded the quantaat each level. This information can be used to determine ifthe quantum assigned to a particular level should be increasedor decreased.

D, TEST AND SET INTERRUPTS

Test and set interrupts are used to protect common dataareas in a multiprocessor configuration. A common data areais o-ne that may be concurrently accessed by more than oneactivity. Each area to be protected has a test and set wordassociated with it. Before accessing a protected area, anactivity executes a test and set (TS) instruction on theprotection word. If the word is clear, it is set; and the

9

access takes place. If the word is set, a test and setinterrupt occurs; and the activity continues to retry untilthe word is cleared.

SIP counts the number of test and set interrupts foreach test and set word. SLEXl , which protects the switchlist, normally has the highest counts. Abnormally high testand set counts indicate processor contention and should beinvestigated

.

10

IV. MEMORY MANAGEMENT

A. MEMORY CONFIGURATION

The maximum memory space available on UNIVAC 1108 's is262K words. Because memory is limited, it assumes increasedimportance when evaluating system performance. Drawing amemory configuration is a useful first step in determininghow efficiently memory is being utilized. The configurationshould show how much memory is reserved for system functionsand how much is available for user programs. The dynamiccharacteristics of memory behavior will determine if it isdistributed in the most efficient manner. The generalmemory configuration for UNIVAC 1108 's is illustrated inFigure IV-1. The dynamic characteristics of memory usageare described in the sections that follow.

B. MEMORY ALLOCATION

User main storage is dynamically controlled within theEXEC by the Dynamic Allocator (DA) . The DA is responsiblefor allocating main storage (core) to all active users inthe system. User requests for core are maintained on theCore Request Queue (CRQ) . The DA selects the highest priorityrequest from the CRQ and attempts to assign memory to it inthe most economical way possible. The desirability ofassigning a particular segment of core to a program is basedon a set of internal cost factors.

When programs are initially loaded into main memory,they are assigned a core quantum. A core quantum representsa guaranteed amount of time that a program is allowed touse memory. Once assigned a core quantum, a programcannot be forced to be swapped out by a higher priorityactivity until its core quantum has expired. Programs thathave completed their core quantum are eligible to be swappedout to make room for other activities.

Gaining access to memory does not guarantee access tothe CPU. An activity in memory must compete with all otheractivities in memory on a pure priority basis for access tothe CPU. CPU access is also controlled by quanta; however,unlike core quanta, CPU quanta do not guarantee that anactivity will not be interrupted by a higher priorityactivity.

Situations exist wherein a program may be eligible forswap out even though its core quantum has not expired. Themost common situation arises when the program is waiting for

11

Resident EXEC I-Bank

Permanent Segment Area

EXEC Segment Overlay Area

User Main Storage

EXPOOL Expansion

Resident EXEC D-Bank

EXPOOL

Interrupt locations,permanently residentEXEC instructions

Transient segments selectedfor permanent residency

Area reserved for executionof transient segments

User programs and dataEXEC segments on a spaceavailable basis

Expansion area if EXPOOLbecomes tight

Permanently residentEXEC data storage

EXEC data storage poolBuffers are requested andreleased as needed

MEMORY CONFIGURATION

FIGURE IV-1

12

the completion of Symbiont I/O operation (READ$ or,PRINT$).In general, whenever a program enters a long wait state, itbecomes a candidate for swap out.

Three performance hypotheses arise in connection withmemory allocation:

(1) Inefficiently assigned core priorities prevent thesystem from properly discriminating among users.

(2) Core quanta that are set either too low or too highare causing unnecessary system overhead in the form ofexcessive swapping.

(3) Fragmented memory is causing blocks of memory to gounused and programs to remain swapped out longer thannecessary

.

1 . Core Priorities

Access to main memory is controlled by core priorities.Two sets of core priorities are defined: in-core prioritiesand request priorities. That is, a program may have adifferent core priority while it is waiting on the CRQ thanwhen it is actually loaded into memory. The two sets ofpriorities are summarized in Table IV-1 . The followingobservations about Table IV-1 are noteworthy:

(1) Both demand and batch programs have an in-corepriority of 07 when they are using their initial quantum.This guarantees that they will be loaded before allother programs that have already received at least onequantum.

(2) The core request priority of 10 is reserved forbatch programs that have been elevated by the DynamicAllocator Periodic Adjustment Routine (DAPA) to satisfythe demand/batch sharing algorithm.

(3) Batch programs have a higher in-core priority thanrequest-core priority. This assures that waiting batchprograms will not cause in-core batch programs to beswapped.

(4) EXEC segments residing in user program area havelower priorities than all user programs. Since EXECsegments may reside in user program space, this guar-antees that user programs can get the space back v/henthey need it.

13

(5) Common banks have higher priorities than most otherusers

.

This is significant, since system performancemay be improved by configuring certain programs ascommon banks

.

PRIORITY REQUEST-CORE PRIORITIES IN-CORE PRIORITIES

00 UNUSED RESIDENT EXEC.

01 UNUSED DOWNED MEMORY02 REAL-TIME REAL-TIME03 COMMON STATIC BANKS COMMON STATIC BANKS04 TIP PROGRAMS TIP PROGRAMS05 CRITICAL DEADLINE CRITICAL DEADLINE06 INCORE MOVE REQUEST COMMON DYNAMIC BANKS07 COMMON DYNAMIC BANKS INITIAL QUANTUM PROG10 ELEVATED BATCH PROG DEMAND

11 -17 DEMAND DEMAND20 -27 UNUSED BATCH/NON-CRIT DL30 -37 BATCH/NON-CRIT DL UNUSED40-50 UNUSED WRONG MEMORY TYPE

51 UNUSED LONG-WAIT BANKS52 UNUSED LOAD-IN-PROGRESS SEG53 UNUSED ACTIVE SEGMENTS54 UNUSED WAITING SEGMENTS55 UNUSED INACTIVE SEGMENTS

56 -76 UNUSED UNUSED77 UNUSED AVAILABLE CORE

CORE PRIORITIES

TABLE IV-1

The problems associated with demand and batch corepriorities are described below.

a. Demand Core Priorities . Demand core priorities area function of program size and the parameter CQFACT. IfCQFACT is incorrectly specified, the system will notproperly discriminate among demand users. The demandcore priority (CQHL) is computed as

CQHL = 010"'' + Min (7, L)

This report follows the UNIVAC convention of writingoctal integers with an initial 0.

14

where L is the program level. The program level is afunction of program size S and C=CQFACT , as given by thefollowing equation:

L = S^+C+C/2 s2

The brackets denote that only the integer portion of theresult is used. The program size S is specified in coreblocks (512 words)

.

To determine the range of program sizes that correspondsto different levels, the above equation can be solvedfor S:

S = v^C (L-3/2)

Since EXEC Level 32R2 currently has C set equal to 800,

S = 20/ 2L-3

The equation for S can be used to generate Table IV-2.

The table demonstrates two important points. First,all demand programs greater than 34K have the same leveland consequently the same core priority. Second, therange of program sizes assigned to a given level decreasesas the level increases. That is, the level equation ismore sensitive to size changes in large programs than itis to size changes in small programs.

LEVELL

SIZE (BLOCKS)S

SIZE (K) RANGE (K)

1 0-19 0- 9 9

2 20-34 10-17 7

3 35-44 18-22 4

4 45-52 23-26 3

5 53-59 27-29 2

6 60-66 30-33 3

7 67- 34-

DEMAND PRIORITY LEVELS FOR CQFACT = 800

TABLE IV-2

15

The range of program sizes assigned to a uniquelevel can be increased by properly adjusting CQFACT. Todouble the range of programs assigned different levels,CQFACT must be increased by a factor of 4. Letting C =4*800 = 3200 produces Table IV-3. When CQFACT = 3200,the system distinguishes among programs in the range 0

to 67. As will be seen shortly, the core priority alsoaffects the computation of core quanta.

LEVEL SIZE (BLOCKS) SIZE (K) RANGE (K)

1 0- 39 0-19 192 40- 69 20-34 143 70- 89 35-44 9

4 . 90-105 45-52 7

5 106-119 53-59 6

6 120-132 60-66 6

7 133- 67-

DEMAND PRIORITY LEVELS FOR CQFACT = 32 00

TABLE IV-3

The Dynamic Allocator Response Time Report from SIPmay be used to determine if CQFACT is properly set.This report specifies the number of requests of eachcore priority that the DA received. If the requests arenot distributed across all priorities, CQFACT shouldprobably be adjusted.

b. Batten Coj.g Priorities . Batch core priorities are afunction of the run card priority. As was noted earlier,batch programs have different core priorities, dependingon whether they are waiting to get in core or whetherthey are already in core. The computation of batch corepriority is similar to that for demand programs:

In core: CQHL = 020 + LWaiting for Core: CQHL = 030 + L

where L is the priority level. The priority level iscomputed directly from the run card priority as follows:

^ _ ( P -'A' ) *8L - ^ ^

^

where P denotes the octal representation of the Priority P.

16

If the equation is evaluated for all values of P, TableIV-4 can be produced.

KUJN LAKU FKxUKlii T X?'\7TPTLltiV tiLl

a R r* n nu

E,F,G 1

H,I,J 2

K,L,M -3

N,0,P,Q 4

R,S,T 5

U,V,W 6

X,Y,Z 7

BATCH PRIORITY LEVELS

TABLE IV-4

Table IV-4 suggests that batch users should beassigned run priorities according to their relativeimportance. If all users default to the same priority,the potential benefit of being able to distinguishbetween different batch users is obviated. Data about,how run card priorities are used may be extracted fromthe Master Log File.

2 . Core Quanta

Quantum size has important performance implications. Ifquanta are set too long, demand response time may be lengthened.If quanta are set too short, excessive swapping overhead maybe generated. The first step in determining how core quantashould be assigned is to understand how they are computed.

The core quantum of a program is a function of (1) theRUN CARD priority, (2) the time required to swap the program,(3) the parameter CQAN, and (4) the core priority. Theobjective is to give the program at least as much coreresidency time as it takes to swap the program in and out.This is an attempt to cut down on swapping overhead. Morespecifically, the core quantum is computed from the followingformula

:

* * S * SFRATE + SFLATECQHL-CLTAL

17

The core quanta are measured in Standard Units of Processing(sup's). One SUP corresponds to a 200-microsecond increment.Hereafter, a 200-microsecond increment is r^^erred to as atic. The maximum quantum size allowed is 2 -1 tics =

262,14 3 tics = 52.4 seconds. The individual terms in thecore quantum formula are described below.

a. Swap Time . The basis of the core quantum calculationis the time required to swap a program. The time toswap a program in and out is given by

S * SFRATE + SFLATE

where S is the program size in core blocks (512 words)

,

SFRATE is the time in tics required to transfer twoblocks to the SWAP$FILE device, and SFLATE is the averagenumber of tics required to make two accesses to theSWAP$FILE device.

The other terms in the formula are adjustments tothe swap time.

b. Run Card Priority . The run card priority affectsthe core quantum multiplicatively . The highest run cardpriority effectively doubles the quantum size. Theadjustment is given by

CZ' + 25 - P)/25

where P equal's the priority from the run card and 'Z' =

037 = 31. For example, if P = 'S', the factor becomes( 31+25-24 ) /25 = 1.28. The minimum and maximum valuesare ( 31+25-31 ) /25 = 1.0 and (31+25-6)/25 = 2.0,respectively.

c. CQAN Adjustment . CQAN is the core quantum dynamicadjustment factor. It is currently hardcoded at 15.Since it is divided by 10 in the quantum equation, itincreases the core quantum 1.5 times. CQAN is the onlyexternal parameter available for adjusting the corequantum size calculation.

d. Core Priority . When a program exceeds (uses up) itscore quantum, its core priority is reduced; that is, itspriority level is increased by one, up to a maximumof 7. The next time the program is loaded, its quantumsize is doubled by multiplying the quantum by

I

CQHL-CLTAL

|

2

18

where CQHL is the highest priority allowed for theprogram and CLTAL is the priolrity last assigned to theprogram. The computation of CQHL is described in thesection on core priorities (IV.B.l.a). CLTAL is givenby

CLTAL = CQHL + N

where N is the number of quanta that have been used sincethe program began executing. Since the maximum spreadbetween CQHL and CL'gAL is 7, tl^e core priority adjustmentlies in the range 2 =1 and 2 = 128.

The fact that CQHL is used to compute CLTAL demon-strates the importance of correctly specifying theparameter CQFACT. If CQFACT is set too low, largeprograms will receive a high (low numeric value) initialcore priority. This means that the program will havemore opportunities to double its quantum size. Systemperformance could be affected if many large programsbegin doubling their core quanta.

e. Example . Consider the following parameter values:

P = 'S' = 030 = 24

S = 14 blocks = 7K

CQAN =15

SFRATE = 33 tics/block for 1782

SFLATE = 220 tics for 1782

CQHL = Oil = 9

CLTAL = 017 = 15

The core quantum Q is given by

Q = 2f^"-'-^l * [ (31 +25-24 ) ^ 15^ * 14 * 33 + 220]25 10

= 70851 tics

= 14.2 seconds

19

f. Problem Detection . Performance problems associatedwith core quanta may be detected from SIP. The DAResponse Time Report specifies the number of requeststhat exceeded their initial core quantum. If thesenumbers are high, it may be advantageous to increaseCQAN. Performance for important batch users may beimproved by properly assigning run card priorities.

3 . Core Fragmentation

As memory is allocated and released, empty spaces arecreated in memory that are too small to hold any availableprogram. This memory space goes unused, and memory becomesfragmented. The Memory Fragmentation Report from SIP reportthe number of program loads that failed because of memoryfragmentation. If memory fragmentation is found to be aproblem, it may be necessary to make adjustments to thememory cost allocation table (CTABL)

.

CTABL is a set of seven parameters the DA uses tooptimize the allocation of memory. The cost table isdefined with two conflicting objectives in mind: (1) maxi-mize memory utilization and (2) minimize the amount ofswapping. Swapping would be eliminated if a program wereallowed to remain in memory until it completed execution;however, this would result in very poor memory utilization.The cost table attempts to achieve a balance between thesetwo objectives.

The DA allocates memory in two phases: the first phase(Phase 0) attempts to allocate memory based on a bank'spreference for primary or extended storage. Banks areconsidered in the following order: require primary, requireextended, prefer primary, prefer extended, and no preferenceThe second phase (Phase 1) searches for available memory onan individual, bank-by-bank basis.

A separate set of cost factors is used for each phase.Of the seven factors in the cost tables, three are formaximizing memory utilization and four are for minimizingswapping. A sample cost table is given in Table IV-5.Each factor in the cost table is used as a cost or weight toperform a particular allocation. If C. is the cost forfactor i and X. is the value of factor i, the cost to makea particular allocation is given by

7

Allocation Cost = E C . *X

.

i=l ^ ^

20

Phase 0 cost factors attempt to minimize swapping, whilePhase 1 cost factors attempt to maximize memory utilization.

PHASE 0 PHASE 1

COST FACTOR COST COST

Core Priority 1 1

Distance from Edge 32 1, 024Blocks not used , 1 2, 047Blocks with I/D Conflict 1,024 32Blocks not i^ PreferredStorage Type

^

512 8

Blocks to swap out^

1,024 64Programs to suspend 640 64

Minimize Swapping

COST TABLE

TABLE IV-

5

C. DEMAND/BATCH SHARING

If system resources are not being properly distributedbetween demand and batch users, it may be necessary toadjust the parameters associated with the demand/batchsharing algorithm. The demand/batch sharing algorithm isdriven by the three parameters DMIN, DMAX , and DINC. DMINspecifies the minimum percentage of core block SUP's (CBSUP's)guaranteed to demand users. DMAX is used to compute themaximum percentage of CBSUP's that may be used by demand pro-grams. That is, DMAX specifies the minimum percentage ofCBSUP's guaranteed to batch programs.

The DA Periodic Adjustment Routine (DAPA) is responsiblefor guaranteeing that the required percentages are beingsatisfied. DAPA computes the maximum allowed demand percent-age as

MAXDEM = Min (DMAX, DMIN + DINC * DOPEN)

where DOPEN is the number of demand runs open at a particularinstant and DINC is the incremental percentage added to DMIM

21

for each demand run open. MAXDEM varies dynamically as thenumber of demand runs open varies. MAXDEM is, however,subject to an upper bound of DMAX. The entire process isillustrated in Figure IV-2.

100%

BATCH MAY SWAP DEMANDMAXIMUMPERCENTAGEALLOWEDfor dejiand

deadline may swap demanddemand may swap batch

minim™percentageguaranteedto demand

demand may swap deadline

0

DEMAND /BATCH SHARING ALGORITHM

FIGURE IV-2

MAXDEM

NOPxMAL

:

DMIN

DAPA computes the CBSUP ' s consumed by all demand pro-grams during the previous n seconds . It then computes whatpercentage this is of the total CBSUP 's available duringthat time period. If the derived percentage is larger thanDMIN and less than MAXDEM, no adjustment is necessary. Ifthe derived percentage is greater than MAXDEM, DAPA adjuststhe core priorities of batch programs so that they are higherthan demand programs. This will enable batch programs toswap out demand programs and consume a larger percentage ofthe computer resources available. Similarly, if the derivedpercentage is less than DMIN, demand programs are not receiv-ing their guaranteed percentage. This is corrected by raisingthe priority of demand programs above deadline batch priority.SIP may be used to determine how the parameters are set.The System Idle Activity Report lists the current values forDMIN, DMAX, and DING. It also specifies the average number

22

of demand runs open (DOPEN) . These data can be used todetermine if DINC is reasonably set. For example, DINCshould be set such that

DMIN + DINC * AVG DOPEN < DMAX

The Master Log File can be used to determine the numberof CBSUP ' s being accumulated by demand and batch programs.An estimate can be made from SIP by looking at the averageprogram size and the percent of CPU time consumed for eachtype.

D. FUNCTION/SEGMENT ACTIVITY

The main performance problems associated with function/segment activity are the delays caused by excessive segmentloads. EXEC segments are transient routines that normallyreside on mass storage (disk or drum) . If an activityrequests the service of a segment that is not in core, thatactivity must wait until the needed segment is loaded.Excessive loading of high frequency segments may signifi-cantly affect system performance.

The Function/Segment Activity Report in SIP may be usedto identify which segments are potential bottlenecks. Ofthose segments with the highest request counts, the segmentswith the highest load counts should be investigated in moredetail. Three hypotheses must be tested:

(1) The wrong segments have been made permanentlyresident.

(2) The size of the segment overlay area is incorrectlyspecified.

(3) The sticking powers assigned to the segments areincorrectly specified.

These problems are discussed in the following sections.

1 . Resident Segments

The most direct approach to minimizing segment loads isto make the most active segments permanently resident. Thisapproach, however, should be reserved for the most criticalsituations. For example, if response time delays could be

23

directly related to the time required to load a particularsegment, that segment should be considered a candidate forpermanent residency. In many installations, the segmentD5CRIT satisfies this requirement.

Segments may be made permanently resident by specifyingthem in an XXXSEG statement at system generation time. TheXXXSEG statement specifies how EXEC segments are to beconfigured in the system.

2 . Segment Overlay Area

The next most direct approach for minimizing segmentloads is to increase the amount of storage permanentlyreserved for EXEC segments. When memory is initiallyconfigured, the EXEC reserves a fixed block of storage forthe execution of EXEC segments. Unless otherwise directed,the EXEC reserves a space equal to the size of the largestsegment. If the size of the largest segment is 6K, thesegment overlay area may hold as many as 12 1-block (512-word)segments. The size of the segment overlay area may easilybe increased by properly specifying the parameter OVSIZE.

Increasing the size of the overlay area does not guaranteethat certain segments will always be resident; it onlyincreases the probability that they will be resident. If asegment has a high sticking power (core priority) , theprobability will be even higher.

Segments do not have to execute in the segment overlayarea, and they may (and commonly do) execute in the userprogram area. However, when executing in user core, theyhave a lower core priority than any user program. If thespace they occupy is required by a user program, they willbe overlaid. (EXEC segments are reentrant and are neverswapped out. They are simply overlaid and reloaded whenneeded again)

.

3 . Sticking Power

"Sticking power" is another name for a segment's corepriority. The term indicates the importance of retaining acopy of the segment in core. Sticking power is a four-digitoctal number in the form SSFF, where SS is the segmentstatus and FF is the segment frequency. The order of thenumbers determines that status is the most significantcomponent of sticking power. A segment's status changes

24

dynamically as the state of the segment changes. The fourpossible states and associated status values are summarizedin Table IV-6. Note how these correspond to the corepriorities in Table IV-1.

STATUS STATE

52 Load-In-Progress (LIP)53 Active54 Waiting55 Inactive

SEGMENT STATUS VALUES

TABLE IV-6

Frequency is the only part of sticking power that can beexplicitly specified. Frequency is defined at systemgeneration time by the parameter FREQ in the element AAPCT.Frequency is an estimate of the relative expected usage of aparticular segment. Since segment usage is very workloaddependent, the default values may be incorrect for a partic-ular installation.

Proper frequency values are easy to assign. The segmentswith the highest request counts should have the highestfrequencies. Increasing a segment's frequency should in-crease the probability of it being in core when it is requested.

When changing segment frequency, the following idiosyn-crasy should be noted. Since frequency is a component ofcore priority, low values correspond to high priorities. Inthe element AAPCT, just the opposite is true. High valuescorrespond to high priorities. Frequencies in the range00-77 may be specified.

E. EXPOOL

EXPOOL, the Executive Storage Pool, is a block of mainstorage reserved for use by the EXEC. EXPOOL is organizedinto a set of buffers that are acquired and released asrequired. All buffers are initially one core block (512

I

words )^in length; however, the EXEC may request buffers ofsize 2 where n may range from 2 to 9 . When the EXEC

25

requests a buffer smaller than any that already exists, alarger buffer must be successively halved until the requiredsize is available. For example, a request for a four-wordEXPOOL buffer may result in a 512 word buffer being halvedseven times.

The major performance problems associated with EXPOOLare listed below:

(1) EXPOOL size is set either too high or too low.

(2) Not enough EXPOOL expansion blocks are reserved.

(3) EXPOOL thresholds are improperly specified.

Each of these problems is discussed in the following sections.

1. EXPOOL Size

The size of EXPOOL is computed at system generation timeby the formula

EXPOOLSIZE = (e P . *W. +EXPADJ) * (9/8)i

where P. is the value of parameter i and W. is the number ofwords oi EXPOOL to be configured for each unit value of param-eter i. For example, consider the number of EXPOOL wordsrequired to support the maximum allowable number of batch runs.MAXOPN specifies the maximum number of batch runs allowedopen, and EXPBCH estimates the number of EXPOOL words neededto support each batch run. Therefore, MAXOPN*EXPBCH wordsof EXPOOL will be reserved to support batch runs. Similarrequirements are computed in the element E8END for approxi-mately 20 other parameters.

Since the EXPOOL requirement computed from the parametersis an estimate, the computed value may have to be adjusted.EXPADJ is a configuration parameter for this purpose. Forextra safety, a 12.5% cushion is added to the final value.This is done by multiplying the final value by 9/8. A finaladjustment, not shown in the equation, is to round theEXPOOL size up to an integral multiple of 01000.

Changes to EXPOOL size are made by changes to EXPADJ.The proper size is easily determined by monitoring theEXPOOL Activity Report in SIP. This report specifies themaximum number of EXPOOL words in use during the monitoringperiod. If this number is continually less than the number

26

of EXPOOL words allocated (also reported by SIP) , EXPADJshould be set to a lower value. This may mean a largernegative number. This is particularly important on aUNIVAC 1108, wherein memory is a critical resource.

The reverse problem is also important; i.e., not enoughEXPOOL is allocated. Such a condition will show up as anEXPOOL tight condition or an EXPOOL expansion. In theextreme case, all EXPOOL expansion blocks will be used, andthe system will crash with an 070 stop. These problems arediscussed in the next two sections.

Before assuming that EXPOOL size is too low, potentialproblem areas should be investigated. The most common causeof EXPOOL being used up is a program that uses an inordinatenumber of mass storage granule tables. Another common causeis excessive operator key-ins. Both these conditions uselarge numbers of EXPOOL words.

2. EXPOOL Expansion

Whenever an EXPOOL request is made and there are onlytwo core blocks of EXPOOL space remaining, EXPOOL willexpand. EXPOOL expands one core block at a time, up to amaximum of EXPEXP blocks. EXPEXP is a CONFIG parameter thatspecifies the maximum number of 512-word expansion blocksallowed. Because EXPOOL expands so that the EXPOOL arearemains contiguous (see the configuration diagram in FigureIV-1), user programs may have to be swapped out so thatmemory can be reorganized. Once EXPOOL expands, it does notcontract

.

EXPEXP is an important parameter. If the system usesall the expansion blocks available and still needs more, thesystem will crash with an 070 stop. EXPEXP should thereforebe set high enough to avoid such a catastrophe.

3. EXPOOL Thresholds

Before EXPOOL thresholds can be explained, the prioritiesassociated with EXPOOL requests must be described. Inaddition to specifying the size of EXPOOL buffer required,each EXPOOL request has a priority associated with it.There are two main priority types: non-immediate and imme-diate. Non-immediate requests are non-critical requeststhat may be queued if no EXPOOL buffers are available.Alternatively, non-immediate requests can specify an abnormal

27

return rather than be queued to wait for an availablebuffer. Immediate requests, on the other hand, are crit-ical. They must be satisfied immediately, and they cannotbe queued. If they cannot be satisfied, the system willcrash (070 stop)

.

Non-immediate requests are further divided into low,medium, and high priorities. If no priority is specified,medium priority is assumed. SIP summarizes the number ofEXPOOL requests by both size and priority.

Associated with each non-immediate priority is a threshold.The purpose of each threshold is to guarantee that a certainnumber of EXPOOL words will be reserved for the next higherpriority request. The thresholds are EXTLOW, EXTMED, andEXTHIG for priorities low, medium, and high, respectively.When the EXPOOL in use reaches one of the thresholds, allrequests with a priority less than or equal to the priorityfor that threshold will be queued. For example, when theEXPOOL in use reaches EXTHIG, only immediate requests willbe honored. All non-immediate requests will be queued orgiven abnormal returns.

Whenever EXPOOL in use reaches one of the thresholds,EXPOOL is said to be "tight." SIP reports the number oftimes EXPOOL became tight and the number of requests thatoccurred while EXPOOL was tight.

EXTLOW, EXTMED, and EXTHIG are hardcoded in the elementE8END. If they are not specified properly, EXPOOL may notbecome tight soon enough to prevent a crash; or EXPOOL maybecome tight too soon and cause severe system performancedegradation. They are specified as octal percentages ofEXPOOLSIZE. For example, the default value for EXTHIG isEXPOOLSIZE * 076/0100. The corresponding multipliers forEXTLOW and EXTMED are 070/0100 and 074/0100, respectively.

Another threshold used to signal a shortage of EXPOOLspace is EXSCH. EXSCH is defined as EXTLOW-EXPBCH , whereEXPBCH is the number of EXPOOL words required to support onebatch run. If the EXPOOL in use goes above EXSCH, theCourse Scheduler (CSN) prevents any additional batch runsfrom being opened.

28

V. FACILITIES MANAGEMENT

A. MASTER FILE DIRECTORY

The Master File Directory (MFD) contains the controlinformation needed to manage catalogued files. At the heartof the MFD is the directory look-up table. The look-uptable contains a one-word pointer to the detail informationfor each catalogued file in the system. The look-up tableis indexed by a hash code computed from the file name. Iftwo file names are similar, there is a chance they will havethe same hash code and point to the same entry in the look-up table. If this happens, the look-up table entry must bemodified to point to a search item. The search item containspointers to the detail information for each file with thathash code.

Four performance hypotheses apply to the efficient useof the MFD:

(1) Too small a look-up table is causing duplicate hashcodes.

(2) Many similar (near duplicate) file names are causingduplicate hash codes.

(3) Activities attempting to search the MFD at the sametime are being delayed because of contention.

(4) The MFD is improperly located on the mass storagedevices

.

1 . Look-up Table Size

The smaller the size of the directory look-up table, thehigher the probability of duplicate hash codes. Duplicatehash codes have two adverse effects: (1) the MFD takeslonger to search, because an additional level of pointers isintroduced that requires an additional I/O operation; and(2) mass storage space is wasted because a 28-word searchitem is generated for as few as two file names with the samehash code.

The size of the look-up table may be controlled by theconfiguration parameter DCLUTS. UNIVAC recommends DCLUTS beassigned a prime number at least twice as large as the

29

number of catalogued files in the system. Using a primenumber reduces the probability of two file names having thesame hash code.

2 . Similar File Names

Similar file names increase the probability of duplicatehash codes. This may be minimized in two ways: (1) use aprime number for DCLUTS and (2) avoid naming conventions thatrepeat major portions of a file name.

A file's hash code is an index to the file's appropriatelook-up table entry. This hash code is obtained from ahashing algorithm that operates on the file's file name andqualifier. The file name and qualifier are combined by thehashing algorithm to form a single 18-bit half-word. Thishalf-word is divided by the parameter DCLUTS, which speci-fies the look-up table's length. The remainder from thisdivision is the index into the look-up table. The entrycorresponding to the file name/qualifier combination isfound by adding the remainder to the look-up table's begin-ning address. Similar file names could therefore causeseveral files to have the same entry in the look-up table.

3 . MFD Contention

Access to the MFD look-up table is controlled by a setof lock cells. The number of lock cells is computed fromthe parameter DLOKSF:

where DLOKSF must be an integer in the range 0 to 4 . N,therefore, lies in the range 1 to 16.

The MFD is divided into N equal parts, and each part iscontrolled by a separate lock cell. When an activityaccesses the MFD, only the portion of the MFD that containsits file is locked. The remainder of the MFD is availablefor concurrent use by another activity. If DLOKSF is set to0, no concurrent accessing of the MFD may take place.

It would at first appear that DLOKSF should always beset to 4. This is not true. When the EXEC needs to lockthe entire MFD, it must set each cell individually. Increas-ing DLOKSF by 1 doubles the overhead associated with lockingthe entire MFD.

30

4. MFD Location

The MFD look-up table should be located on the fastestmass storage device available (usually a drum) . The MFDdetail information for each file is contained on the samedevice on which the file resides. This information com-prises a device directory. For most mass storage devices,the first directory track begins in Position 0. Threeexceptions to this rule are FASTRAND II, FASTRAND III, and8460 disks.

The parameters POSDIR and POS60 control the positioningof the directory on FASTRAND and 8460 devices, respectively.To minimize arm and head movement, the parameters should beset so that the directory is located near the middle of theallocated file space. Alternatively, if one or two filesaccount for the majority of a device's activity, the direc-tory should be located near those files.

B. MASS STORAGE ROLLOUTS

The EXEC manages mass storage as a virtual resource.That is, from a user's viewpoint, an unlimited amount ofmass storage is available. When more mass storage is re-quested than is physically available, the EXEC releasesspace by rolling out infrequently used files to magnetictape

.

The rollout process is controlled by the five thresholds,MSWl through MSW5 . These thresholds, which specify criticallevels of mass storage in tracks, are computed at systemgeneration time (see Table V-1) . When the total number ofunallocated mass storage tracks drops below MSWl, massstorage is said to be "tight." This tight condition isdetected by the facilities allocation element FALL. When atight condition is detected, FALL calls the element ROLOUT,which actually initiates the unload/rollout operation.

THRESHOLD DEFINITION TYPICAL VALUE

MSWl MSW2+MSW4 1664 tracksMSW2 MSW3+MSW4 1248 tracksj^MSW3 0150*MAXOPN 832 tracksMSW4 MSW3/2 416 tracksMSW5 0200 128 tracksMSTOL MSW4/4 104 tracks

Assumes that MAXOPN = 8

.

TABLE V-1: MASS STORAGE THRESHOLDS

31

Separate actions are triggered as unallocated massstorage falls below each thresholds An actual rollout tomagnetic tape may be initiated when MSWl is passedc MSW2simply causes a warning message to be printed on the opera-tor's console. When MSW3 is passed, ROLOUT places a CoarseScheduler Hold on all demand and batch runSc MSW4 generatesanother warning message. Finally, if mass storage decreasesbelow MSW5

, any user requesting mass storage for file expan-sion is terminated. However, any EXEC mass storage requestsare still honored.

The physical rollout process is not performed by theelement ROLOUT, ROLOUT starts a predefined (canned) runstream, which in turn calls the SECURE Processor, TheSECURE Processor is responsible for the management of allcatalogued files in the system, and rollout is one of itsfunctions

„

Four performance hypotheses are associated with the massstorage rollout process:

(1) Mass storage rollout is occurring either too soon ortoo late,

(2) Frequent rollouts are causing unnecessary systemoverhead,

(3) Since an excessive amount of mass storage is rolledout during each rollout operation, frequent rollbacksare occurring,

(4) The SECURE processor is unloading the wrong filesand allowing inactive files to occupy mass storage,

1 „ Rollout Initiation

As described above, the rollout process is based on thethree threshold values MSWl, MSW3 , and MSW5 , Setting thesevalues involves a trade-off. If they are set too high, therollout process will be initiated while a significant amountof available mass storage remains. If they are set too low,the rollout process will be delayed, and mass storage requestsby user programs may be terminated,

MSWl is the most important threshold because it mayinitiate an actual rollout of mass storage to tape (see page34). MSWl should be set according to the system's fileallocation activity. This activity is based on (1) thenumber of runs entering the system and (2) the mass storagedynamic expansion by existing runs. An additional considera-tion is the amount of time it takes the operator to mount

32

the rollout tape. If the file allocation activity causesmass storage to decrease below the MSW5 threshold before arollout completes , MSWl should be increased. However, MSWlshould not be so large as to cause the rollout of massstorage with a significant amount of remaining space. Thiswould cause a portion of mass storage to remain unused.

The MSW3 threshold is important for two reasons. First,MSWl, MSW2 , and MSW4 are currently computed as a function ofMSW3 (see Table V-1). Thus, alterations to MSW3 must con-sider the effect on the other thresholds. Second, MSW3determines when a Coarse Scheduler Hold will be placed ondemand and batch runs. MSW3 should be set according to thedynamic expansion activity of the currently open runs. Ifthis activity causes mass storage to decrease below MSW5

,

MSW3 should be increased. As a guideline, MSW3 should bethe average file space required for a run multiplied by theaverage number of runs concurrently active.

Whenever mass storage is reduced to the MSW5 threshold,only EXEC mass storage requests are statisfied. MSW5 shouldtherefore be set large enough to handle these requests.

The principal information source concerning the optimalsetting of MSWl, MSW3 , and MSW5 is the console warningmessages . A unique warning message is displayed when avail-able mass storage falls below each threshold. The rate atwhich these messages appear serves as a guideline for settinthe thresholds . If the MSWl warning message is quicklyfollowed by the MSW2 and MSW3 messages, the MSWl thresholdshould be increased. If the MSW3 message is quickly followeby the MSW4 and MSW5 messages, MSW3 should be increased.The MSW3 message should be displayed only occasionally; theMSW5 message should almost never be displayed.

2 . Frequent Rollouts

Frequent mass storage rollouts might occur for tworeasons: (1) too few tracks are being unloaded during eachrollout and (2) a large number of mass storage requestsoccurs within a short time period. Although little can bedone about the second reason, the first can be controlled.

The element ROLOUT calculates the number of tracks to beunloaded from the following formula:

NBTRK = (f^) - AV

33

where TC is the total mass storage capacity in tracks con-figured in the system, AV is the number of available (unal-located) tracks at the time of the rollout, and PERCNT is aparameter that specifies the fraction of total system massstorage tracks to be made available as result of a rollout.For example, if PERCNT = 20, l/20th (5%) of all configuredmass storage should be available after the rollout iscomplete. Since AV tracks are currently available, theamount to be unloaded is reduced by this amount. The numberof tracks to be unloaded is constrained so that

MSWl <_ NBTRK < 030000

That is, at least MSWl tracks, but no more than 030000 tracks,are unloaded for each rollout. This constrained value ofNBTRK is plugged into a SECURE Processor command to unloadmass storage. The command as it appears in the predefinedrun stream is

UNLOAD TRACKS = NBTRK.

Unless the number of tracks (NBTRK) is specified, the UNLOADcommand releases 3000 tracks.

The SECURE Processor first attempts to satisfy the UNLOADcommand by eliminating files which have current copies onmagnetic tape. Since these files are already duplicated ontape they are just marked in the MFD as being unloaded andno I/O to tape is done. As a last resort, the SECUREProcessor will select files without current copies on tapesfor unloading. To unload these files, I/O is required totransfer them to magnetic tape.

The number of rollouts that occurs during a certaintime period may be obtained from the Master Log File. SECUREperforms the rollout operation under the RUNID ROLOUT. IfMaster Log Data is unavailable, the SIP I/O Trace report maybe used to obtain the approximate number of rollouts thatoccurred. This report specifies the number of words unloadedfor each file. Converting the number of words to tracks andsumming them will give the total number of tracks unloaded.Dividing the total number of tracks by NBTRK will yield theapproximate number of rollouts

„

3 . Frequent Rollbacks

Frequent rollbacks may occur because too much massstorage is unloaded during each rollout. This problem iseasily resolved by increasing PERCNT so that fewer tracksare unloaded each time.

34

4 . Unload Eligibility Factor

The unload eligibility factor (UEF) determines whichfiles are to be unloaded during a rollout operation. If theUEF is improperly specified, files may have to be reloaded(rolled back) shortly after they are rolled out. The SECUREProcessor computes a UEF for each catalogued file. Thosefiles with the highest UEF are unloaded first.

The UEF is a function of six factors: (1) time sincelast reference, (2) mean time between references, (3) size,(4) equipment type, (5) F-Cycle, and (6) type: public orprivate. Of these factors, only the first three are signifi-cant. The UEF is computed from the following formula:

UEF = CURNCY*P (TSLR) + FREQCY*P (MTBR) + SI ZEWT*P ( SI ZE

)

where TSLR equals time since last reference, MTBR equals

-

mean time between references, and SIZE equals file size intracks. The notation P (FACTOR) denotes the number of pointsassigned to the specified factor. CURNCY, FREQCY, and SIZEWTare weights associated with each factor.

The point values assigned to each factor are obtained bya table look-up operation. The table used to look-up thepoints for TSLR and MTBR is given in Table V-2. The pointvalues for SIZE are given in Table V-3. The UEF is normal-ized so that it lies in the range 0 to 63. Three UEF valuesare reserved for special use. A UEF of 0 is reserved forfiles that are already unloaded; a UEF of 1 is reserved forcatalogued tape files and removeable disk files; and a UEFof 2 is reserved for files that are "unload inhibit" (V option)

.

The optimal point values and weights depend on theunique file characteristics of an installation. The charac-teristics of each file may be extracted from the MFD andsummarized by the program LISTER. In release 19R1 Version 2

of the SECURE Processor, CURNCY is set to zero. This impliesthat TSLR has no effect on which files are selected forunloading. The other weights have a value of 1 in the samerelease Df SECURE.

The calculation of UEF by SECURE is illustrated by thefollowing example. The file ACOB*PFPROJECT has the follow-ing characteristics:

TSLR = 4 days 15 hours = 4.63 days

Time since catalogued = 104 days

Number of references = 46

35

HOURS DAYS POINTS HOURS DAYS POINTS

1X 1X ft n 1X oo

VJ • u o o ^ o 4 n ft1 . u o 1 7X /

U • -L ^ X X o 4 ft ^ 1 ftX oA A*± X J 1 ^ ft 1 QX y

7 X ^ ^

1 n o 1 ft?X o ^ 7 i^iR/.JO ^ X

-L -J 7 X ^ ft ft ^

16 0. 67 8 242 10.08 2321 0 88 9 272 11 ^3 2428 1.17 10 302 12.58 2534 1.42 11 344 14.33 2640 1.67 12 386 16 .08 2750 2. 08 13 428 17.83 2860 2.50 14 470 19.58 2970 2. 97 15 512 21.33 30

544 23.08 31

TABLE V-2: POINT VALUES FOR TSLR AND MTBR

TRACKS POINTS TRACKS POINTS

20 1 500 1640 2 550 1760 3 600 1880 4 650 19

100 5 700 20120 6 810 21140 7 920 22160 8 1030 23180 9 1140 24200 10 1250 25250 11 1360 26300 12 1470 27350 13 1580 28400 14 1690 29450 15 1800 30

1900 31

TABLE V-3: POINT VALUES FOR SIZE

36

MTBR = 2.26 days

SIZE = 2 tracks

From Tables V-2 and V-3, the point values are

P (TSLR) = 18

P (MTBR) = 14

P (SIZE) 1

Therefore, the UEF is 0*18+1*14+1*1 = 15.

C. GRANULE TABLES

Mass storage is allocated in increments called granules.Granules may be specified in either tracks or positions.One track equals 64 sectors or 1792 words; one positionequals 64 tracks or 114,688 words. The EXEC sets up bufferscalled granule tables to keep track of the physical massstorage assigned to each file. Granule tables contain 26-wordentries. Each entry in the granule table reserves a physi-cal block (granule) of mass storage. If track granularityis specified, each word in the granule table represents 1792words of mass storage. If position granularity is specified,each word in the granule table represents 114,688 words ofmass storage. If a file occupies non-contiguous mass storagespace, the intervening space must also be included in thegranule tables. Granule tables are stored on mass storageand brought into EXPOOL buffers when they are needed.

Four performance hypotheses apply to granule table usage:

(1) Users are specifying the wrong file granularity ontheir assign cards.

(2) The maximum number of granule tables allowed inEXPOOL when EXPOOL becomes tight is inefficientlyspecified.

(3) The maximum number of mass storage sectors requestedat one time for file granule tables is too small.

(4) The default values for the maximum number of granulesassigned to a file is excessive.

37

1 . File Granularity

A file is allocated mass storage in granules. Thus, ifa file with position granularity uses only one track of aposition, the entire position is allocated to that file. Acomplete track is allocated to a file when track granularityis specified.

Specifying track granularity utilizes mass storage moreeffectively than position granularity. If a file with trackgranularity uses only part of a track, only a portion of atrack is unused. However, if position granularity is speci-fied for the same file, 63 additional tracks would be allo-cated to the file. These additional 63 tracks will beunused and unavailable for allocation to other files. Sincetrack granularity results in a finer allocation of massstorage, it is usually more efficient than position granulari

Position granularity should be specified only for large,contiguous files. The main advantage of position granularityis a reduction in main storage usage for PCT and EXPOOLbuffers. Position granularity uses a single granule or v/ordof memory to specify 64 contiguous tracks. For the same 64tracks, track granularity will use 64 granules or words ofmemory. There is therefore a reduction in the amount ofmain storage used. The swap file is one case wherein posi-tion granularity will reduce the amount of main storageneeded for granule tables.

2 . Granule Tables in EXPOOL

A file's granule tables, which are located in thedirectory of the file's device, are chained together.Whenever part of a file is accessed, the granule tablereferencing that part is brought into EXPOOL. If an I/Ooperation references a granule not addressed by the granuletable in EXPOOL, the other granule tables are read in one ata time until the proper granule table is found.

The parameter NGTB specifies the maximum number of afile's granule tables permitted in memory when EXPOOL becomestight (when the remaining number of EXPOOL buffers hasdropped below a critical threshold) . If a file has morethan NGTB granule tables when EXPOOL becomes tight, granuletables are paged out until only NGTB remain.

38

Two performance problems are associated with NGTB

.

First, if NGTB is set too large, valuable EXPOOL space isoccupied by excess granule tables. Second, if NGTB is settoo small, excessive granule table paging may result. As anexample, suppose NGTB were set to one and a random accessfile were alternatively referencing two granule tables.Each reference would cause one granule table to be writtenout and the other to be read in. If NGTB were equal to 2

and both granule tables were in EXPOOL, the paging would beeliminated.

NGTB may be determined by deciding the amount of EXPOOLthat should be reserved for granule tables when EXPOOL istight. If AMT denotes this amount, NGTB may be found by thefollowing formula:

where NFILE is the number of concurrent files using granuletables and 032 represents the number of words per granuletable

.

3 . Granule Table Mass Storage Sectors

File granule tables are maintained in the device directory,and each granule table occupies one mass storage sector.The hardcoded parameter MAXSEC specifies the maximum numberof granule tables (mass storage sectors) that may be requestedby the EXEC at one time. The EXEC element FALL determineswhether the number of requested sectors is greater thanMAXSEC. If the requested number is greater, MAXSEC is used.MAXSEC is also used to determine the required EXPOOL buffersize for the element FASEC. (FASEC allocates sectors ofmass storage space for the EXEC.)

The main problem associated with JIAXSEC is setting ittoo small. If a request is for more sectors than I4AXSEC

,

the remaining sectors will be requested again. Overhead isgenerated because multiple FASEC requests occur when obtain-ing the required number of mass storage sectors. MAXSECshould be set to the 80th or 90th percentile of the mostnumber of granule tables needed to describe the largest file.

4 . Default Value for Granules

An option on the facility control statements allows theuser to specify the maximum allowable file length in granules.

39

When this value is unspecified, MAXGRN is substituted.MAXGRN is a configuration parameter that defines the defaultvalue for a file's maximum number of granules. MAXGRN orthe specified value is used to indicate when the file'slength exceeds the maximum numiDer of granules.

The primary reason for specifying maximum file length isto prevent a user from accidentally overallocating his filespace. Because this is more likely to occur during thedebug phase of a program, a possible value for MAXGRN is the80th percentile of all file sizes.

D. MASS STORAGE MANAGEMENT PRACTICES

Improvement in facility performance is obtainablethrough mass storage management. An overall review of massstorage utilization should be made before the facilityparameters are altered. This would eliminate most of thesimple mass storage problems. Since mass storage usage isdynamic, the review should be made on a periodic basis.Depending on the system activity, a biweekly or monthlyreview may be appropriate.

Three performance hypotheses apply to mass storagemanagement

:

(1) Incorrect file placement is causing excessive I/Otimes and unit contention.

(2) Non-contiguously allocated mass storage file spaceis causing excessive I/O and unused mass storage.

(3) Mass storage space is being wasted by unused cata-loged files.

1 . File Placement

Several factors must be considered when attempting tooptimize file placement: the size of the file, the frequencyof file access, the potential for file contention, and theperformance characteristics of the available mass storagedevices. In general, the most frequently accessed filesshould be placed on the fastest devices. If the files arelarge or the devices are small, this may not be reasonable.Similarly, if many frequently accessed files are placed onthe same device, a device contention bottleneck may result.

40

The following guidelines may be helpful when attemptingto optimally place files:

(a) Place the most frequently accessed files (usuallyEXEC files) on the fastest device (usually a drum)

.

Contention and its resulting queuing may have to betolerated if no more high speed devices are available.

(b) Distribute large, frequently accessed files acrossseveral devices to minimize the possible contention.

(c) Assure that files are allocated contiguous space onthe devices to which they are assigned.

(d) Distribute frequently accessed files across severaldevices of the same speed.

(e) If the controller is found to be the source of mostqueuing (rather than the individual devices), considerinstalling a dual access controller.

The information needed to make the above decisions maybe obtained from a combination of SIP, MFD map (DIRVER) , andMFD list (LISTER) . The Main Storage Analysis by Unit Reportin SIP summarizes the queuing associated with the controllersand the units. DIRVER may be used to identify the largestfiles; LISTER may be used to identify the most frequentlyreferenced files.

2 . Non-contiguous Mass Storage Allocations

Non-contiguous mass storage occurs when mass storage isallocated in clusters throughout a device. This causes anewly catalogued file to be allocated in pieces throughoutthe device. Non-contiguous allocations cause an increase inthe number of seeks required to read or write a file. As aresult, the I/O times for this file will be greater than fora contiguously allocated file. There will also be an increasein overhead because of the multiple-parting (M-part) of I/Orequests, i.e., the user makes I/O requests for a group oflogically contiguous words that are actually non-contiguouson mass storage. The EXEC must compute additional accesscontrol words (ACW's) in order to handle these non-contiguousblocks

.

The most convenient indicator of whether a system's massstorage space is allocated non-contiguously is the number of

41

multiple-parting I/O requests that occur. The SIP reportthat deals with the number of times various EXEC areas wereentered contains this information. This report provides thenumber of times the EXEC area dealing with M-parts was entered.If this number is large, non-contiguously allocated massstorage is indicated. Specific, large, frequently accessedcatalogued files with non-contiguous allocations can beidentified by using (1) a MFD list (LISTER) for the numberof accesses since the file was catalogued and (2) a MFD map(DIRVER) for determining if the files were non-contiguous.

Two actions may alleviate non-contiguous mass storagespace. First, unused catalogued files should be deleted (asdiscussed in the following section) . This will make moremass storage available and increase the probability of afile being allocated contiguously. Second, the file adminis-trator should periodically review how large, frequently usedfiles are allocated. If a file is identified as beingnon-contiguously allocated, it should be transferred to tapeand rewritten to a contiguous mass storage area. To make alarge enough contiguous space, several smaller files mayalso need reallocation. The device that the file is beingallocated to should be considered at the same time that thefile is reallocated. If the file is accessed frequently,assigning the file to a higher speed device may beadvantageous

.

3 . Unused Catalogued Files

One possible reason for mass storage unavailability isthe abundance of unused, mass storage catalogued files. Agood user practice is to delete a catalogued file when it isno longer needed. If the users always delete their unusedcatalogued files, (1) more mass storage space is available,(2) there is less chance of allocating mass storage non-contiguously, and (3) mass storage rollouts will be reduced.If there are users who do not delete their unused files,then a procedure such as archiving might be employed.Archiving is the deletion from mass storage of all cataloguedfiles not referenced for a fixed time period. These deletedfiles are transferred to tape. At the same time the file isdeleted from mass storage, the file's Master File Directoryentry should also be eliminated. This will reduce thenumber of unnecessary directory items. For example, cata-logued files not accessed for seven days might be archivedto tape at the end of each week. Ideally, archiving willprevent unnecessary EXEC initiated rollouts when mass storageallocation activity is highest. The archiving processshould be done during non-production hours.

42

Another possible reason for mass storage unavailabilityis that users are creating catalogued files instead 'of usingtemporary files. This causes the " same situation as wouldthe unused catalogued files. The users should create acatalogued file only when it is necessary.

E. SCHEDULING

The Course Scheduler handles the EXEC schedulingfunctions. Batch runs are placed into a schedule queueafter the legality of the (SRUN control statement is checked.Demand runs are immediately opened. When either a batch ordemand run is opened, the Course Scheduler reads the run'sfirst set of facility control statements. If the requestedfacilities are unavailable, the run is placed in a facili-ties hold queue. When the run's facilities are available,the run is fully accepted and the Dynamic Allocator is givencontrol

=

Two performance hypotheses apply to scheduling:

(1) The number of demand and batch runs simultaneouslyopen does not effectively utilize computer resources.

(2) The installation is not taking advantage of the runcard priorities to assist in scheduling work.

1 . Demand/Batch Runs Open

A system's run mix can consist of demand and batch runs.The emphasis that the Course Scheduler places on either runtype should be based on the objective of the computersystem. If the objective is service to demand users, careshould be taken to see that performance is not degraded bybatch runs. However, the number of batch runs allowedsimultaneously open should be set high enough to utilizethose computer resources incompletely used by demand runs.For example, batch runs may use the CPU while demand runsare waiting for I/O to complete.

The two parameters that affect the balance betweendemand and batch are CDHOLD and MAXOPN. CDHOLD specifiesthe maximum possible number of simultaneously active demandruns. Similarly, MAXOPN specifies the maximum possiblenumber of batch runs open at one time. CDHOLD should be setto allow the greatest number of active demand runs without

43

(1) degrading demand response time to an unacceptable levelor (2) preventing the minimum required batch throughput.MAXOPN should then be set high enough to allow the greatestnumber of batch runs to be simultaneously open withoutdegrading demand response time. If MAXOPN is set as largeas possible, the Course Scheduler will control the computerutilization. The Course Scheduler will select from thebacklog those batch runs that will use the remaining avail-able computer resources.

Data needed for setting CDHOLD and MAXOPN are obtainedfrom two SIP reports. The System Idle Activity report givesthe average number of demand and batch runs open; theResponse Time Report provides the demand response time.Demand response time is an indicator of the service thatusers are obtaining. If the average number of open demandruns is close to CDHOLD and demand response time is small,CDHOLD can be increased. The same can be done for batch byadjusting MAXOPN. If demand response time becomes unaccept-able, CDHOLD and/or MAXOPN can be reduced until satisfactoryservice is obtained (assuming some other condition is not atfault)

.

2 . Run Priorities

The proper use of run priorities will result in (1)

providing equitable system availability to all users and (2)

establishing a priority structure that will allow criticalruns to receive better service. Whenever possible, userruns should be assigned a priority. This allows the discrim-ination of various types of runs. For example, a distinctionshould be made between a run's degree of I/O- versus CPU-proneness. An I/O-prone run should receive a higher prioritythan a CPU prone run. Of course, the urgency of the run isanother factor to be considered.

MPRIOR is one of two configuration parameters thataffect run priorities. MPRIOR specifies the maximum allowedpriority on a (SRUN statement. If the @RUN card priority ishigher than MPRIOR, then MPRIOR is substituted. MPRIOR isuseful in determining the number of priority classes thatmay be assigned to a run. Since the possible priorities are"A" through 'Z" (with "A" being a higher priority than "Z"),up to 26 different classes (A through Z) are possible. IfMPRIOR is set to R, only nine classes are possible.

44

APRIOR is the second configuration parameter thataffects run priorities. APRIOR is the default priority usedfor @RUN statements with no priority specified. Assumingthat most runs have a priority assigned, APRIOR should beset to a lower priority than the majority of runs withpriorities assigned. However, if most runs default forpriority, the purpose of a priority structure is defeated.

The number of runs at each priority is obtainable fromthe Master Log File. A report could be written that summa-rizes the general characteristics of those runs at eachpriority. From this report, it may be determined if thecurrent priority structure is satisfactory or if runs withcommon characteristics need assignment to a different runpriority.

45

VI . INPUT/OUTPUT

A. LATE ACKNOWLEDGMENTS

A late acknowledgment is the failure of the CPU torespond to a control unit within a critical response time.Every time a word is transferred between the CPU and a

control unit, the CPU must respond to a signal from thecontrol unit. For example, when a word is ready for input,the control unit activates its Input Data Request (IDR)

line. When the word has been received by the CPU, the CPUsends an Input Acknowledge (lA) signal back to the controlunit acknowledging receipt. The control unit can acceptanother word from the device only after the lA signal hasbeen received from the CPU. Otherwise, the previous wordmight be over-written before it has been accepted by theCPU. During data output, the CPU must respond to an OutputData Request (ODR) signal with an Output Acknowledge (OA)

signal within a critical response time. If a late acknowl-edgment occurs, the next word is not transferred until thedevice's next revolution.

Four performance hypotheses apply to late acknowledgments

:

(1) Non-optimal drum interlace factors are resultingin slow effective transfer rates.

(2) Scatter read and gather write requests on a disksubsystem are causing persistent late acknowledgments.

(3) Symbiont scatter/gather I/O is causing lateacknowledgments

.

(4) Late acknowledgments are occurring on a memorymodule overloaded by I/O requests.

1 . Interlace Factors

Five possible interlace factors are used on the FH-432,FH-1782, and FH-880 drums: 1, 2, 4, 8, and 16. An inter-lace of 1 implies that logically consecutive data words arestored in contiguous drum locations. If the drum interlaceis n, n-1 drum locations are skipped between consecutivedata words. Doubling the interlace factor halves the drumtransfer rate. Interlace factors may be altered either byresetting a switch or by physically changing several inte-grated circuit boards within the drum's controller. The

47

configuration parameters DHLACE, DHLACH, and DHLAC8 tell theEXEC what interlace factors are being used for the FH-4 32,FH-1782, and FH-880 drums, respectively. If they are incor-rectly specified, the EXEC vill erroneously compute timinginformation associated with the drums.

The optimum interlace factor involves a trade-offbetween obtaining the highest data transfer rate and thenumber of late acknowledgments that occur. Whenever a lateacknowledgement occurs, an additional drum revolution(called a missed revolution) is required to read or writethe next word. The number of late acknowledgments mighttherefore be great enough to offset the doubling of thetransfer rate by using a lower interlace factor. The opti-mum interlace factor should be the least possible value of1, 2, 4, 8, and 16, where the number of late acknowledgmentswill not decrease the effective transfer rate below theslower transfer rate obtainable at a higher interlace factor.

The SIP Input/Output Activity Report is useful in deter-mining the optimum interlace factor. The following infor-mation is obtained from this report:

TIME = Total channel busy time in seconds

WORDS = Number of words transferred

ACW = Number of I/O operations

AVAXST = Time per I/O operation in millisecondsexcluding data transfer time

AVAXST includes both the access time and the time lost dueto missed drum revolutions. The number of missed drumrevolutions N may be estimated by

_ ACW * (AVAXST-A)

This formula depends on two important assumptions thatmay not always be true: (1) the number of ACW's equals thenumber of I/O operations and (2) the true average accesstime is equal to the theoretical. The first assumption isnot true for scatter/gather I/O, and the second is not truefor a sequence of I/O operations to the same area on drum.Each subsequent I/O request will have missed the next dataword and will have to wait another complete revolution toaccess the word.

43

where A equals the average access time of the drum and Dequals the delay per missed drum revolution. The averageaccess times for the FH-432 is 4.3 ms and the FH-1782 is17.0 ms. The delay times for a missed revolution are 8.5ms for the FH-432 and 34.0 ms for the FH-1782.

The effective transfer rate ER (including the misseddrum revolutions) is found by the following formula:

ER = W/ (W/R+N*D)

In this formula, W equals the number of words transferred(as recorded by SIP) and R is the maximum transfer rate atthe present interlace. The factor (W/R) is the time totransfer W words at R words per second; the factor (N*D) isthe time lost due to missed drum revolutions. If theeffective transfer rate ER is less than the transfer rateobtained at the next higher interlace factor, the higherfactor should be used. If ER is greater than the nexthigher interlace factor's transfer rate, the present inter-lace factor is satisfactory. Another, similar experimentshould be performed at the next lower interlace factor todetermine if a higher transfer rate is possible on the drum.

2 . Scatter/Gather I/O Requests

Late acknowledgments may occur on a disk subsystemduring scatter read and gather write requests. A scatterread is an I/O technique wherein a contiguous block of datathat resides on a peripheral device is read into multiple,non-contiguous main storage buffers. These data are trans-ferred in a single, continuous I/O operation. A gatherwrite is the converse I/O operation--writing from severalmain storage areas to one mass storage area.

An I/O operation uses an access control word (ACW) tospecify the length and location of each I/O buffer in mainstorage. During a scatter/gather request, the EXEC mustswitch control between ACW s as the I/O operation transfersdata to or from each buffer. This switching of ACW s is apotential cause of late acknowledgments on a disk subsystemwith a high transfer rate. Since the I/O operation iscontinuous, ACW s must be switched within the time intervalbetween the transfer of consecutive data words.

Late acknowledgments may be minimized by the configura-tion parameter INCORE . INCORE applies only to disk sub-systems. When INCORE is set to one, code is generated to

49

I

ensure that each ACW specifies a main storage buffer 'whoselength is an integer multiple of a" disk's record length.This guarantees that ACW switching will occur on recordboundaries. When ACW' s are switched on a record boundary,more time is available for changing ACW's because additionaltime is required to pass over the interrecord gap.

The SIP Input/Output Activity Report may be used todetermine if late acknowledgments are occurring on a disksubsystem. The following information is obtained from thisreport

:

ACW = Number of I/O operations

AVAXST = Time per I/O operation in millisecondsexcluding data transfer time

The number of late acknowledgements N may be estimatedby

N = ACW* (AVAXST - A)D

Since AVAXST includes both the access time and time lost dueto missed disk revolutions, the disk's average access timeA is subtracted from AVAXST. The value D equals the delayper missed disk revolution. If scatter/gather requests areused often and N is large, INCORE should be set to 1.

As general guidelines, UNIVAC recommends that INCORE beset to 1 on (1) any 1106 system or (2) an 1108 system with8440 disks configured. If persistent late acknowledgesoccur on an 1108 or 1106 II system, INCORE should also beset to 1. INCORE should be set to 0 on (1) 1108 systemsconfigured with 8414, 8424, or 8425 disks on a high prioritychannel or (2) any 1110 system.

3 . Symbiont Scatter/Gather Requests

Late acknowledgements may also occur during symbiontscatter/gather I/O requests. The symbiont complex maintainsits data in chains of 28-word buffers within core. Duringan I/O operation, symbiont data are read into or writtenfrom this chain of eight buffers. A total of 224 words istransferred at one time. These words are transferred directlyto or from these eight buffers during a scatter/gather I/O.Since the I/O operation is continuous, the EXEC must be ableto change buffers between the transfer of consecutive words.If the EXEC cannot switch to the next buffer fast enough, alate acknowledgment occurs.

50

Symbiont late acknowledgments may be reduced by settingthe configuration parameter SCGA to 0. Symbiont scatter/gather is disabled when SCGA equals 0. Instead of trans-ferring data directly to or from the eight 28-word buffers,a contiguous 224-word EXPOOL buffer is used. During a writeoperation, the data of the eight 28-word buffers are trans-ferred into a 224-word EXPOOL buffer. The actual I/O opera-tion will then write the data from the 224-v7ord buffer to aperipheral device. Conversely, the data read from a periph-eral device will be transferred into the 224-word buffer andthen later dispersed into a chain of eight 28-word buffers.

UNIVAC states that SCGA must equal 0 if U20 tapes existon an 1106, 1108, 1108A or 1100/20 system. The U20 tapesare not supported because of their high transfer rate ofabout 53,33 3 words per second.

4 . Memory Module Overloading

Late acknowledgments, also called data overruns, mayoccur when a memory module becomes saturated by I/O accesses.This will occur when a memory module's transfer rate is lessthan the combined transfer rate of the subsystems active onthe memory module. The number of late acknowledgments (dataoverruns) can be reduced by setting the configuration param-eter lOLM to 1. lOLM equated to 1 causes the assembly ofcode for the EXEC's I/O Load Monitor.

I/O Load Monitor reduces late acknowledgments by adjust-ing the I/O activity on memory modules that are nearlysaturated with I/O requests. I/O Load Monitor does this bymaintaining the available I/O transfer rate (LMLVL) for eachmemory module. LMLVL is the module's maximum I/O transferrate with the sum of subsystem transfer rates currentlyactive on the module subtracted out. Before a request goesto the device handler, I/O Load Monitor compares this addi-tional subsystem's transfer rate with LMLVL of the request'smemory module. If the request's subsystem transfe: rate isgreater, the request is queued. If not, the request'ssubsystem transfer rate is subtracted from LMLVL, and therequest goes to the device handler. When a request's I/Ocompletes, the memory module's LMLVL is increased by theamount of the completing request's subsystem transfer rate.Another adjustment is made to LMLVL, depending on whether adata overrun occurred during the completed request's I/O.If an overrun occurred, LMLVL is decreased by at least 12.5%of LMLVL. This will reduce the number of future data over-runs. If the request completed normally and a queue exists,LMLVL is increased by at least 1.56% of LMLVL.

51

Information about the number of late acknowledgments(data overruns) that are occurring can be obtained from theSIP report I/O Load Monitor Activity. If nonrecoverabledata overruns are occurring^ the percentages used to adjusta memory module's LMLVL might be altered.

lOLM may be set to 0 (turned off) when the sum of thesubsystem transfer rates is less than a memory module'smaximum I/O transfer rate. For example, a system could haveup to five FH-1782 drums transferring data to one memorymodule before the sum of the subsystem transfer rates wouldequal a memory module's maximum transfer rate. However, ifdata overruns begin to occur, the monitor should be turnedon (lOLM = 1)

.

B. SEEK TIME

The time to access a disk record is the sum of seek timeand latency time. Seek time is the time needed to move thedisk's read/write head to the required cylinder. Latencytime is the time the disk rotates before the desired positioncomes under the read/write head. Under pre-seeking, theEXEC initiates disk seeks for requests that are queued.When those requests are serviced, the disk v^ill be on positionand no seek time will be required.

Four performance hypotheses apply to pre-seeking:

(1) Pre-seeking is disabled on 8440 disk drives.

(2) The parameters used for selecting the next I/Orequests for pre-seeking are set wrong.

(3) Angular addressing on the 8440 disk subsystem isdisabled

.

(4) Full disk pack seeks are degrading disk performance.

1 , 8440 Disk Pre-seeking

When UNIVAC first introduced the 8440 disks, a hardwareerror was associated with the 8440 pre-seeking function. Asa result, the parameter NOSEEK was placed in the element 10.This parameter prevented 8440 pre-seeking until the problemwas corrected. Currently, NOSEEK is equated to 1-STPAUL.If NOSEEK equals 1 (STPAUL = 0) , 8440 pre-seeking is disabled.NOSEEK equated to 0 (STPAUL = 1) enables 8440 pre-seeking.

52

The hardware error associated with 844 0 pre-seeking wascorrected by Field Change Order (FCO) Number 037 (applied tothe disk controller). Field Support Bulletin Number 252,dated 22 August 1974, details the application of pre-seekingto the 8440 disks. If a site has this hardware error cor-rected, then NOSEEK should be equated to 0 in order toenable pre-seeking. This problem does not apply to suchother disks as the 8414 's and 8460 's.

The amount of improvement obtained by pre-seeking dependson the computer system's configuration and the system'sworkload. Pre-seeking should provide significant improve-ment on an 8440 disk subsystem if (1) there are more thantwo disks per control unit or more than three disks per dualaccess controller, (2) the workload is I/O-prone, and (3)the file activity is distributed evenly between disk unitsand throughout each disk unit. The SIP Input-Output Activityreport is useful for determining whether pre-seeking isadvantageous. Pre-seeking should decrease the percent oftime a disk subsystem's channel is busy. The value ofAVAXST should also decrease (the average time per I/O opera-tion, excluding data transfer time)

.

2 . On-Position Requests

The EXEC element 10 attempts to minimize disk accesstime (read/write head movement) by servicing on-positionrequests before other requests. An on-position request isa request that references the cylinder where the read/writehead of the disk is currently positioned. The parameterMAXOPR limits the number of on-position I/O requests thatwill be serviced consecutively. After MAXOPR on-positionrequests have been serviced, the disk's position is advancedone position. The disk I/O handler moves the read/writehead in one direction until all requests in that directionhave been checked. When an edge is reached, the scan direc-tion is reversed. If no on-position requests are found whensearching the request queue, the request closest to thecurrent disk position (in the scan direction) is selectedfor service. Since the disk only scans in one direction ata time, it is possible for the previous on-position requestto be closest and again given service.

To prevent a request from being permanently locked outby on-position requests, the parameter MAXLLC is used.MAXLLC specifies the maximum number of times that a requestmay be passed over. A request is considered passed over if,after having been selected as the closest request, it is

53

then rejected because either an on-position request or acloser request was found. A request may be rejected onlyonce each time the request queue is searched. When 10detects a request that has been rejected flAXLLC times, thatrequest is serviced immediately.

MAXLLC should be set larger than MAXOPR; otherwise, thepurpose of MAXOPR will be defeated. The value of MAXOPRshould be based on the length of time that requests may belocked out because of on-position requests » The time alloweddepends on whether the computer's workload is demand orbatch. If the workload is demand, a series of on-positionrequests should be allowed to monopolize a device for only ashort period of time (say, one-half second) . In a batchenvironment, the potential lockout time may be longer (say,one second) . After the allowed lockout time is established,MAXOPR can be calculated from the SIP Input-Output Activityreport. For each disk subsystem, the average I/O time perrequest is obtained by dividing the channel busy time by thenumber of ACW's. MAXOPR for the subsystem is then obtainedby dividing the allowed lockout time by the average I/O timeper request. If several disk subsystems exist, potentialvalues of JIAXOPR should be calculated for each subsystem.The actual value assigned to MAXOPR should be based on themost active subsystem.

For example, assume the maximum allowed lockout time ina demand environment is established at 0.5 seconds. FromSIP, the following data may be obtained:

8414 Channel Busy Time = 567,544 seconds

Number of 8414 ACW's = 17,119

Assuming one I/O request per ACW, the average I/O timeper request is 0.033 seconds. MAXOPR is calculated to be 15requests (0.5/0.033).

3 . Angular Addressing

Angular addressing, also called rotational positionsensing, is another feature of the 8440 disk subsystem.Like pre-seeking. Field Change Order (FCO) Number 037 mustbe implemented on the 5033 Control Unit before the 8440disks may use angular addressing. Angular addressing iscontrolled by the parameter ANGLOF.

54

when an I/O request is initiated, a command is given tothe control unit specifying an angular position on a track.The control unit will then determine if the disk is morethan a given distance away from the addressed position. Ifit is, the channel is logically disconnected from the opera-tion. Just before the disk rotates to the position wherethe read or write will occur, the channel is logicallyreconnected and the read or write occurs. This assumes thatthe channel was not busy transferring another unit's datawhen reconnection was supposed to take place. If the channelwas busy, the procedure is repeated during subsequent diskrevolutions. In these cases, the service time will actuallybe longer than it would be without angular positioning.

4 . Full Pack Seeks

A performance problem occurs if a disk pack has frequentlyused files allocated at the extremes of the pack. A simpli-fied example is one file located on the pack's outer cylinderand another file located on the inner cylinder. This wouldgenerate full pack seeks and would result in an increase inaccess times.

A special case of this problem is treated by the configu-ration parameter FASTSK. When the EXEC initially allocatesdisk space to a file, the inner cylinder is allocated first.Once the inner cylinder is full, the EXEC returns to theouter cylinder to continue allocation. Since this processoften splits a frequently accessed system file, frequentlyfull pack seeks may occur. By setting FASTSK to 1, theinner cylinder is marked allocated, and file fragmentationis prevented.

One disadvantage associated with using FASTSK is areduction in the amount of space available for allocation.If an 8414 disk is prepped at 112 words per record, UNIVACstates that 36 out of 3072 tracks will be unavailable forallocation. A general solution to the seeking problem is toforce frequently accessed files to be placed close together.This would eliminate frequent full pack seeks and not reducethe amount of mass storage available for allocation.

C. BLOCK SIZE

I/O block size is the number of words transferred in oneI/O operation. Block size may degrade performance in eitherof two ways:

55

(1) Block sizes that are not integer multiples Of thephysical disk record size are causing frequent readbefore write operations.

(2) Small average block sizes are resulting in loweffective transfer rates and high EXEC I/O overhead.

Each of these problems are discussed below.

1 . Read Before Writes

Read before write I/O operations are generated by theEXEC when the I/O block size is not an integer multiple ofthe physical record size. Since the smallest unit of I/O isthe physical record size, a block which is smaller than therecord size must be written out as if it were a full block.Since this might destroy good information at the end of thephysical record, the EXEC (1) reads the physical record intocore, (2) changes the part of the record that is beingwritten into, and (3) writes the entire record back to disk.

Read before write operations may be detected by SIP I/OTrace. If a large number of them are occurring, two ap-proaches may be used to reduce their frequency: (1) requireprograms to specify an I/O block size that is an integermultiple of the physical record size or (2) change thephysical record size to agree with frequently used I/O blocksizes

.

In general, disks may be prepped at one, two, or foursectors per record. Since there are 2 8-words per sector,disk record sizes may be 28, 56, or 112 words per record.The only exception to the above rule is that a disk used forEXEC files must be prepped at 28 or 112 words per record.Since most disks are prepped at 112 words per record, I/Oblock sizes should be multiples of 112 words. Since someEXEC I/O is done in 2 8-word blocks, it may be advantageousto prep the system disk at 2 8 words per record.

2 . Small Block Sizes

I/O block size is the number of words transferred in oneI/O operation. When this number is small, the effectivetransfer rate for a large volume of data is substantiallyreduced. This is because access times are typically muchlonger than transfer times. Consider the following example

56

on a 1782 drum with a transfer rate of 240,000 words persecond and an average access time of 17 msec. A file of10,000 words blocked at 100 words per block would require1.74 seconds to transfer. Of this time, only 0.04 secondsis devoted to the actual transfer of data; the remainingtime is due to access time. The same file transferred inblocks of 1000 words would require only 0.21 seconds totransfer, an eightfold improvement.

Besides reducing I/O time, large block sizes substan-tially reduce EXEC overhead. Fewer interrupts are gener-ated, resulting in less I/O processing. The trade-off isthe increased memory size required to hold the buffers.

57

VII. TEST AND DEBUG

The EXEC contains parameters that control internal errorchecking and the maintenance of trace entries. If thesystem is unstable and crashes frequently, these parametersare a valuable asset in debugging the system. Most of thisdebugging information is found in a PANIC dump. On a stablesystem, however, these parameters can be turned off and thenumber of trace entries can be reduced.

Two performance hypotheses that relate to test and debugparameters are:

(1) EXEC instructions are being executed to performunneeded error checking and event tracing.

(2) Critical memory space is being used for storingtrace entries and test and debug program instructions.

Table VII-1 lists the test and debug parameters andidentifies how each parameter adversely affects EXEC over-head. For the purposes of this table, the cutoff for lowand high time is based on the frequency that the code mustbe executed; the cutoff used for low and high memory usageis 200 words. Each of these parameters is described inAppendix B.

PARAMETER

ADVERSE AFFECT ON EXEC OVERHEAD

TIME SPACE

DADBUG Low LowDATRAC High LowEXPTRACE High LowFCDBSZ High HighFNCCHK High LowlODBUG High HighSIPDAT High LowSIPIO High LowSTPAUL High HighSWPCKO High LowUN IVAC Low High

TEST AND DEBUG PARAMETERS

TABLE VII-1

59

APPENDIX A

PERFORMANCE HYPOTHESES

61

I. PROCESSOR MANAGEMENT

1. CPU Priorities . Batch programs that are assignedthe same CPU priority as demand programs are causing demandprograms to have degraded response times.

2. CPU Quanta . CPU quanta that are set either too highor too low are causing the CPU to be inefficiently allocated.

3. Test and Set Interrupts . An excessive number oftest and set interrupts are causing programs to be delayeduntil data areas are unlocked.

II. MEMORY ALLOCATION

1. Core Priorities . Inefficiently assigned core prioritiesprevent the system from properly discriminating among users.

2. Core Quanta . Core quanta that are set either toolow or too high are causing unnecessary system overhead inthe form of excessive swapping.

3. Core Fragmentation . Fragmented memory is causingblocks of memory to go unused and programs remain swappedout longer than necessary.

4. Demand/Batch Sharing . System resources are notbeing properly distributed between demand and batch users.

III. FUNCTION/SEGMENT ACTIVITY

1. Resident Segments . The wrong segments have beenmade peirmanently resident.

2. Segment Overlay Area . The size of the segmentoverlay area is incorrectly specified.

3. Sticking Power . The sticking powers assigned tothe segments are incorrectly specified.

IV. EXPOQL

1. EXPOOL Size . EXPOOL size is set either too high ortoo low.

2. EXPOOL Expansion . Not enough EXPOOL expansionblocks are reserved.

3. EXPOOL Thresholds . EXPOOL thresholds are improperlyspecified.

63

V. MASTER FILE DIRECTORY

1. Look-up Table Size . Too small a look-up table iscausing duplicate hash codes.

2. Similar Filenames . Many similar (near duplicate)filenames are causing duplicate hash codes.

3. MFD Contention . Activities attempting to search theMFD at the same time are being delayed because of contention.

4. MFD Location . The MFD is improperly located on themass storage devices.

VI. MASS STORAGE ROLLOUTS

1. Rollout Initiation . Mass storage rollout is occurringeither too soon or too late.

2. Frequent Rollouts . Frequent rollouts are causingperformance degradation.

3. Frequent Rollbacks . Since an excessive amount ofmass storage is rolled out during each rollout operation,frequent rollbacks are occurring.

4. Unload Eligibility Factor . The SECURE Processor isunloading the wrong files and allowing inactive files tooccupy mass storage.

VII. GRANULE TABLES

1. File Granularity . Users are specifying the wrongfile granularity on their assign cards.

2 . Granule Tables in EXPOQL . The maximum number ofgranule tables allowed in EXPOOL when EXPOOL becomes tightis inefficiently specified.

3. Granule Table Sectors . The maximum number of massstorage sectors requested at one time for file granuletables is too small.

4. Default Value for Granules . The default value forthe maximum number of granules assigned to a file is excessive.

VIII. MASS STORAGE PRACTICES

1. File Placement . Incorrect file placement is causingexcessive I/O times and unit contention.

64

2. Non-Contiguous Mass Storage Allocation . Non-contiguouslyallocated mass storage space is causing excessive I/O andunused mass storage.

3. Unused Catalogued Files . Mass storage space isbeing wasted by unused catalogued files.

IX. SCHEDULING

1. Demand/Batch Runs Open . The number of demand andbatch runs simultaneously open does not effectively utilizecomputer resources.

2. Run Priorities . The installation is not takingadvantage of the run card priorities to assist in schedulingwork.

X. LATE ACKNOWLEDGMENTS

1. Interlace Factors . Non-optimal drum interlacefactors are resulting in slow effective transfer rates.

2. Scatter/Gather I/O Requests . Scatter read andgather write requests on a disk subsystem are causingpersistent late acknowledgments.

3. Symbiont Scatter/Gather Requests . Symbiont scatter/gather I/O is causing late acknowledgments

.

4. Memory Module Overloading . Late acknowledgments areoccurring on a memory module overloaded by I/O requests.

XI. SEEK TIME

1. 8440 Disk Pre-seeking . Pre-seeking is disabled on8440 disk drives.

2. I/O Request Selection . The parameters used forselecting the next I/O requests for pre-seeking are setwrong

.

3. Angular Addressing . Angular addressing on the 84 40disk subsystem is disabled.

4. Full Pack Seeks . Full disk pack seeks are degradingdisk performance.

XII. BLOCK SIZE

1. Read Be fore Writes . Block sizes that are not integermultiples of~The physical disk record size are causing frequentread before write operations.

65

2. Small Block Sizes . Small average block sizes areresulting in low effective transfer rates and high EXEC I/Ooverhead.

XIII. TEST AND DEBUG

1. Execution Time . EXEC instructions are being executedto perform unneeded error checking and event tracing.

2. Memory Space . Critical memory space is being usedfor storing trace entries and test and debug program instruc-tions .

66

APPENDIX B

PERFORMANCE PARAMETERS

67

This appendix summarizes those parameters identified asimportant for tuning system performance. The appendixbegins with a table (Table B-1) that alphabetically listsall the parameters and identifies the EXEC area they affect.Detailed descriptions of each parameter, organized into asection for each EXEC area, follow the table. Each param-eter has the following title line associated with it:

TYPE, ELEMENT, LINE NO, VALUE

where

TYPE = The parameter type: CP ifconfiguration and SW if soft-ware (hardcoded)

ELEMENT = Name of EXEC element wherethe parameter is defined

LINE NO Line number within the elementwhere the parameter is definedon System 7 as of 9 June 1976

VALUE Value on System 7 as of9 June 1976

69

EXEC AREA AFFECTED

c M E F S I Tp E X A c / E

u M P C H 0 S

0 0 I E Tt>K U r

Li u /Y L I U D

T L EI I BE N US G G

PARAMETER C M E F S I T

ANGLOF I

APRIOR SBAWAIT MCDHOLD SCHGDED S

SKSM FCLRCOR MCOR$MAX ECOUNT CCQAN MCQFACT MCTABL MCURNCY F

DADBUG TDATRAC TDCLUTS FDEDLIN S

DHLACE I

DHLACH I

DHLAC8 I

DINC MDLOKSF FDMAX MDMIN MDSWTCHTYP CEXPADJ EEXPEXP E

PERFORMANCE PARAMETERSTABLE B-1

70

EXEC AREA AFFECTED

Ci r co T1m1

P TPb VA A /-<

V- Ei

U •Dr n CO0 0 I E TR 0 L D /Y L I U D

T L EI I BE N US G G

PARAMETER C M E F S I T

EXPRIM EEXPSPL EEXPTRACE TEXSCH EEXTHIG EEXTLOW EEXTMED EFASTSK I

FCDBSZ TFNCCHK TFREQ MFREQCY FINCORE I

lODBUG TlOLM I

LOGEOF FLOGTYP FLWAIT CMAXCRD S

MAXCRQ S

MAXDEM SMAXF4 I

MAXF8 I

MAXF17 I

MAXGRN FMAXLLC FMAXOPN S

PERFORMANCE PARAMETERSTABLE B-1

71

EXEC AREA AFFECTED

c M E F S I Tp E X A. C / E

u M P C H 0 S

0 0 I E TR 0 L D /Y L I U D

T L EI I BE N US G G

PARAMETER C M E F S I T

MAXOPR I

MAXPAG SMAXRTC CMAXSEC FMAXTIM SMAXTUX SMAXTWAIT C

MPRIOR SMSTOL FMSWl F

MSV>?2 FMSW3 FMSW4 FMSW5 FNGTB FNHOLDT SNOSEEK I

OVS I Z E MPERCNT FPMIN MPOOLS IZE I

POSDIR FPOS60 FPRIEXP EQTMTBL CRECBIN ERSINCT S

PERFORMANCE PARAMETERSTABLE B-1

72

EXEC AREA AFFECTED

c M E F S I Tp E X A C / Eu M P C H 0 S

0 0 I E TR 0 L D /Y L I U D

T L E

I I BE N US G G

PARAMETER C M E F S I T

SBLEVl TSBLEV2 TSBLEV3 TSBLEV4 TSBLEV5 TSCGA I

SIPDAT TSIPIO TSIZEWT F

SSDENS FSSEQPT F

STORE MSTPAUL TSWPCKO TSWAP MSYMALL I

SYSPRP FTFEXP FTFMAX FTRMXCO STRMXPO S

TRMXT SUEFTRK FUEFLUP FUNIVAC TWHLDRM FXXXSEG M

PERFORMANCE PARAMETERSTABLE B-1

73

PROCESSOR PARAMETERS

COUNT SW,

DSWTCHTYP SW,

LWAIT SW,

MAXRTC SW,

MAXTWAIT SW,

QTMTBL SW,

DISP, 134, EQU 01000

DAILR, 13, EQU BARH

AAPCT, 3041, EQU 50

DISP, 132, EQU 35000

DISP, 133, EQU 30001

DISP, 3677, See Below

74

PROCESSOR PARAMETERS

SW, DISP, 134, EQU 01000

COUNT specifies the number of words totransfer for the Block Transfer instructionin the Dispatcher IDLE Loop.

SW, DAILR, 13, EQU BATCH

DSWTCHTYP makes the demand switching typeequal to the Batch Type (6) . This was doneto reserve Type 4 for TIP processing. If noTIP processing is being done, DSWTCHTYP maybe reevaluated.

SW, AAPCT, 3041, EQU 50

LWAIT is the length of wait time in milli-seconds. It is used by routines that executean ER TWAIT$ to compute the desired lengthof wait in msec. It is also used by theDispatcher as a threshold to determine ifthe segment long wait bit (SGLWAT) should beset. If the wait time is greater than orequal to (5*LWAIT) , set SGLWAT.

SW, DISP, 132, EQU 35000

MAXRTC is the maximum real time clock valuein 200 ysec increments. The value 35000 isequivalent to 7 sec. IVLAXRTC is used as the"infinite" quantum size for Activity Types 1

through 3

.

SW, DISP, 133, EQU 30001

iyiAXTWAIT is the maximum time wait value inmilliseconds. The value 30001 is equivalentto 30 sec. MAXTWAIT is the longest length oftime an activity will be allowed to waiton the TWAIT queue.

SW, DISP, 3677, See Below

QTMTBL is a 7-word table that holdsthe CPU quanta for Levels 1 through 7 for

75

PROCESSOR PARAMETERS (Cont'd)

Activity Types 4 through 6. The values inmilliseconds for the 1108 are:

LEVEL QUANTUM (msec)

1 1.42 3.03 32.04 64.05 128.06 256.07 512.0

76

MEMORY PARAMETERS

BAWAIT SW, DAPAM, 18, EQU 21

CLRCOR CP, AACONFIG, 7 32, EQU 1

CQAN SW, DA, 17 3, +15

CQFACT SW, AADAEQT, 17, EQU 800

CTABL SW, DA, 280, See Below

DINC CP, AACONFIG, 789, EQU 2

DMAX CP, AACONFIG, 8 02, EQU 50

DMIN CP, AACONFIG, 812, EQU 20

FREQ SW, AAPCT, 4996, See Below

OVSIZE CP, AACONFIG, 1068, EQU 0

PMIN SW, AADAEQT, 21, EQU 5

STORE CP, AACONFIG, 2193, See Below

SWAP CP, AACONFIG, 2903, See Below

XXXSEG CP, AACONFIG, 2.8 91, See Below

77

MEMORY PARAMETERS

BAWAIT SW, DAPAM, 18, EQU 21

CLRCOR

BAWAIT specifies the number of minutes abatch program will be allowed to wait on theCore Request Queue (remain swapped out)before DAPAM will requeue the core requestat a priority higher than all other demandbatch requests (priority 010)

CP, AACONFIG, 732, EQU 1

If CLRCOR is nonzero, core is always clearedbefore loading a program. Core is clearedregardless of the use of the B option atcollection

.

CQAN SW, DA, 173, +15

The core quantum adjustment factor, CQAN, isone of several factors used to compute aprogram initial core quantum. Besides CQAN,core quantum is a function of (1) RUN cardpriority, (2) swap time, and (3) corepriority. The initial core quantum iscomputed by SETCQU in DACCC in 200 ysecincrements

.

1

8

The maximum core quantum allowed is 2 -1 =

262,143 tics = 524 sees (1 tic = 200 ysec).The core quantum is computed from

CQHL-CLTALl 'Z'+25-P *CQAN*SWAPTIME25 10

where CQHL = Highest allowed priorityCLTAL = Current priorityCQAN = Core quantum adjustment factor

P = RUN card priority

CQFACT SW, AADAEQT, 17, EQU 800

CQFACT is used by CQLVL in DACCC and CCRin DA to compute CQHL, the core priority fordemand programs. Demand core priorities liein the range Oil to 017. They are computedfrom

78

MEMORY PARAMETERS (Cont'd)

CQHL = 010 + MIN(7,L)

CTBAL

where L is the program level. The programlevel is a function of program size and C =

CQFACT, as given by the following equation:

L = S'

C+ 1.5

The brackets denote that only the integerportion of the result is used. The programsize is specified in core blocks (512 words).

SW, DA, 280, See Below

The cost table (CTABL) is a set of sevenparameters used to optimize the allocationof memory. The seven parameters are (1)core priority, (2) distance of area from edgeof user space, (3) blocks within area notin use, (4) blocks with I/D bank conflict,(5) blocks not in preferred memory type,(6) blocks that would have to be swappedout, and (7) programs that would have tobe suspended. Each factor X. listed above

associated wit^i it. The costa program to a given area is

has a cost C.of allocatingthen given by

Allocation Cost =E

i=l

C. *X.1 1

The area with the lowest cost is then selectedfor allocation.

DINC

DMAX

DMIN

CP, AACONFIG, 78 9, EQU 2

DC, AACONFIG, 802, EQU 50

CP, AACONFIG, 812, EQU 20

DMIN, DMAX, and DINC are the basicinputs to the demand/batches sharing

79

MEMORY PARAMETERS (Cont'd)

algorithm in DAPA. DMIN specifies theminimum percent of computer timeguaranteed to demand programs . DMAX andDINC are used to compute MAXDEM:

MAXDEM = MIN (DMAX , DMIN+DINC*DOPEN)

where DOPEN is the number of demand runsopen at any instant in time.

MAXDEM specifies the maximum amount oftime allowed for demand programs. Thisguarantees that batch programs willreceive at least (100 -MAXDEM) percentof the system resources.

SW, AAPCT, 4996, See Below

Frequency (FREQ) is a component of segmentsticking power. Sticking power is a four-digit octal number of the form SSFF, whereSS is the segment status and FF is thefrequency. Status may have one of fourvalues, depending on a segment's currentstate: 52 = load-in-progress , 53 = active,54 = waiting, and 55 = inactive. Frequenciesare in the range 00 to 077 and are definedat system generation time in the elementAAPCT. Sticking power is important becauseit is issued as a segment's core priority.

CP, AACONFIG, 1068, EQU 0

OVSIZE defines the function/segment overlaysize. OVSIZE is the number of core blocksdevoted exclusively to transient EXEC segmentsIncreasing this value may increase throughputbut will decrease the amount of core avail-able to user programs. A value of 0 willcause the system to assign the minimumamount of overlay core needed to avoid corelock up, i.e., an amount equal to the sizeof the largest segment.

80

MEMORY PARAMETERS (Cont'd)

SW, AADAEQT, 21, EQU 5

PMIN specifies the number of minutes acommon static bank will be locked in memoryif unused. DAPAr4 scans the EXEC Bank De-scriptor Table (BDT) and marks swappableall common static banks that have not beenreferenced for PMIN minutes.

CP, AACONFIG, 2193, See Below

STORE is a label in a System GenerationStatement (SGS) that defines the mainstorage configuration. It has the form:

STORE FIRST, LENGTH, 'TYPE', INTLV

where

FIRST = First address of areaLENGTH = Length of area in 01000 's

TYPE = Memory type, e.g., '7005' for 1108INTLV = 1 for no interleave and

2 for odd/even interleave

The fields are repeated for each area tobe defined. For example:

STORE 0 , 0100 , • 7005 ', 2

CP, AACONFIG, 2903, See Below

SWAP is a label in an SGS that defines thesizes and allocation of the program swapfile SWAP$FILE. It has the form:

SWAP MAX, 'CRN', 'DEVICE', GRAN

where

MAX = Maximum number of granules thatmay be assigned to the file.

'GRN' = 'TRK' for track granularity' POS ' for position granularity

81

MEMORY PARAMETERS (Cont'd)

•DEVICE' = Any Fastrand formatabledevice type that may be used on an ASGcard, e.g., 'F17' = FH-1782 drum

GRAN = Number of granules to be reservedon this device

' DEVICE ' , GRAN may be repeated as manytimes as needed.

Example: SWAP 25, 'PCS', 'F17', 14, 'F2

' , 1

XXXSEG CP, AACONFIG, 2891, See Below

XXXSEG is a label in an SGS that specifieshow EXEC segments are to be configured instorage. It has the form:

XXXSEG, COND ' SEGl ' , ' SEG2'

, . .

.

where

XXXSEG = XRESEG if segments are to bemade permanently resident for executionin storage.

XXXSEG = XTPSEG if transient segmentsare to be executed in primary storage.Does not apply to 1106/1108.

XXXSEG = XRPSEG if segments are to beexecuted in primary storage. Does notapply to 1106/1108.

COND = An expression which, if itevaluates to zero, indicates that theXXXSEG statement should be ignored.

'SEGn' = A list of segments to which theXXXSEG statement applies.

Example: XRESEG, NOMUX 'DSCRIT'

82

EXPOOL PARAJVIETERS

COR$MAX CP, AACONFIG, 739, EQU 32

EXPADJ CP, AACONFIG, 370, EQU 076000

EXPEXP CP, AACONFIG, 375, EQU 10

EXPRIM CP, AACONFIG, 868, EQU 0

EXPSPL CP, AACONFIG, 889, EQU 'H'

EXSCH sw. E8END, 28 5, EQU EXTLOW-EXPBCH

EXTLOW sw. E8END, 27 5, EQU EXTLNG*070/0100

EXTMED sw. E8END, 27 7, EQU EXTLNG*074/0100

EXTHIG sw. E8END, 27 9, EQU EXTLNG*076/0100

PRIEXP CP, AACONFIG, 1127, EQU 0

RECBIN sw. EXPOOL, 1 9 EQU 0

83

EXPOOL PARAMETERS

CP, AACONFIG, 739, EQU 3 2

COR$iyLAX is the maximum number of log entriesthat is kept in core at any one time. Eachlog entry uses a 32-word EXPOOL buffer. Atrade-off exists between EXPOOL usage andlogging overhead.

CP, AACONFIG, 370, EQU 076000

The EXEC estimates the amount of EXPOOLrequired (EXPREQ) based on (1) the maximumnumber of demand and batch jobs that may beopen simultaneously and (2) the particularhardware configuration. Since EXPREQ mayover or under estimate the amount of EXPOOLactually required, EXPREQ can be adjusted byspecifying an appropriate positive ornegative value for EXPADJ. The number ofwords eventually reserved for EXPOOL isgiven by EXPOOLSIZE and is computed inE8END.

CP, AACONFIG, 375, EQU 10

EXPEXP is the number of 512-word EXPOOLexpansion blocks allowed. EXPOOL does notcontract once an expansion takes place.

CP, AACONFIG, 868, EQU 0

EXPRIM is the percent of EXPOOL in PrimaryStorage (does not apply to 1106/8)

.

CP, AACONFIG, 889, EQU "H"

EXPSPL is the EXPOOL storage priority afterwhich EXPOOL gives buffers in extendedstorage. This allows for fine tuning byshifting allocation among PS and ES (doesnot apply to 1106/8)

.

SW, E8END, 285, EQU EXTLOW-EXPBCH

EXSCH is used to signal to the system thatEXPOOL is approaching a tight condition.EXSCH is equated to the low EXPOOL threshold

84

EXPOOL PARAMETERS (Cont'd)

less the estimated number of EXPOOL wordsrequired for one batch run. EXPBCH is hard-coded in E8END to be 850. EXSCH is used bythe Coarse Scheduler (CSN) to prevent anyadditional runs from being opened, it isused in the UOM change in DAPA to indicatean EXPOOL hold on the operator's console,and it is used by FIMAIN to determine if afacility's hold should be placed on a run.

SW, E8END, 275, EQU EXTLNG * 070/0100

SW, E8END, 277, EQU EXTLNG * 074/0100

SW, E8END, 279, EQU EXTLNG * 076/0100

Threshold values are used to determine whenEXPOOL is tight for the three types of non-immediate priority requests. That is, if alow priority EXPOOL request is made when theEXPOOL storage in use is greater than EXTLOW,EXPOOL is said to be tight for that request,and that request will be queued until theEXPOOL in use drops below EXTLOW. EXTLNG isthe length EXPOOL storage in ExtendedStorage; on the 1108, this is just EXPOOLSIZE.

CP, AACONFIG, 1127, EQU 0

PRIEXP is the number of primary EXPOOLexpansion blocks allowed (does not apply to1106/8) .

SW, EXPOOL, 19, EQU 0

RECBIN is used as a conditional assemblyflag to turn EXPOOL recombination code onand off. Its possible values have thefollowing meaning: 1 = complete recombiningand 0 = do not complete recombining.

85

FACILITIES PARAMETERS

CKbM CP , AACONFIG, 332, EQU 1

CURNLx bW

,

SECURE (MAIN)

,

"TOT _i_ r\16 1, +U

ULLiU i o AACONFIG, 765,

CF , AACONFIG, 793,

CTa7bW / SECURE (MAIN)

,

"7 "5 Q a. 1/ J y , +1

CF , AACONFIG, 488, U U

liUGi Yir LP / AACONFIG, 463, EQU 0

MAA(aKN UF f AACONFIG, 992, EQU 12 8

JMAAbJiC OTa7bW

,

FALL , 8 3, EQU 12 5

Mb i ULi CF , AACONFIG, 1014, EQU MSW4/4

MbWl LP / AACONFIG, 1012,

UF / AACONFIG, 1011, EQU MSW3+MSW4

MbW J CF , AACONFIG, 1009, EQU 0150*MAXOPN

OTa7 y1 LF , AACONFIG, 1010, EQU MSW3/2

JMbWo LP , AACONFIG, 1013, EQU 0200

OTa7bW

,

FALL, 82, EQU 5

n "C^ T)^ IvTrn OTa7bW , ROLOUT, 17, EQU 10

FUbUlK UF , AACONFIG, 1099, EQU 191

FUbD U L.F , AACONFIG, 1107, EQU 135

b J. ^HjW 1 bW , SECURE (MAIN)

,

738, +1

SSDENS CP, AACONFIG, 611, EQU 'V

SSEQPT CP, AACONFIG, 614, EQU '16N'

SYSPRP CP, AACONFIG, 621, EQU 4

86

FACILITIES PARAMETERS (Cont ' d

)

TFEXP CP, AACONFIG, 63 3, EQU 60

TFMAX CP, AACONFIG, 1327, EQU 360

UEFLUP SW, SECURE (SETUP), 2425, See V.B.4

UEFTRK SW, SECURE (SETUP), 2430, See V.B.4

WHLDRM CP, AACONFIG, 65 9, EQU 1

87

FACILITIES PARAMETERS

CP, AACONFIG, 332, EQU 1

CKSM activates the checksum feature of allMaster Bit Tables. The Master Bit Tableassociated with each device will be lengthenedby one word because of the addition of aChecksum Word at the end. CKSM is also usedas a conditional assembly flag for code that(1) adjusts each Master Bit Table's length,(2) generates the checksum, and (3) determinesthe validity of the checksum.

SW, SECURE (MAIN), 737, +0

CURNCY is the weight assigned to time sincelast reference (TSLR) in the computation ofthe unload eligibility factor (UEF) . Aweight of 0 implies that TSLR does notaffect which files are unloaded. Thissetting applies to Level 19R1 VERS 2 ofSECURE.

CP, AACONFIG, 765, EQU 2049

DCLUTS specifies the size in words of theDirectory Loop-Up Table. The DirectoryLook-Up Table is part of the Master FileDirectory (MFD) and is used to index intothe other tables in the directory. The MFDcontains the identification and characteris-tics of each catalogued file.

CP, AACONFIG, 79 3, EQU 2

DLOKSF determines the number of MFD lockcells and is called the MFD lock scalefactor. Each MFD lock cell prevents aportion of the directory look-up table frombeing used. DLOKSF must be an integer from0 to 4 . The number of MFD lock cells (one36 bit word per cell) configured will be:

^DLOKSF

88

FACILITIES PARAMETERS (Cont'd)

Thus, the number of lock cells can rangefrom 1 to 16.

SW, SECURE (MAIN), 739, +1

FREQCY is the weight assigned to mean timebetween references (MTBR) in the computationof the unload eligibility factor UEF. Thissetting applies to Level 19R1 VERS 2 ofSECURE

.

CP, AACONFIG, 488, EQU 0

LOGEOF, in conjunction with LOGTYP, controlsthe conditional assembly of code that specifaction taken on log tape during autorecoveryLOGEOF is valid only when LOGTYP equals 1.If LOGEOF equals 0, an EOF will not bewritten; and the log tape will not be re-wound on autorecovery. The EXEC will con-tinue writing on the extended tape. IfLOGEOF equals 1, an EOF will be written onthe extended log tape; a rewind will bedone; and the next F-cycle assigned.

CP, AACONFIG, 463, EQU 0

LOGTYP controls conditional assembly flags.Code is assembled, depending on whether theMaster Log is written on a FASTRAND deviceor on tape. LOGTYP equals 0 specifies thatthe Master Log is written on a FASTRANDdevice. LOGTYP equal to 1 specifies thatthe Master Log is to be written on tape.LOGTYP is also used in conjunction withLOGEOF to specify action taken during auto-recovery .

CP, AACONFIG, 992, EQU 128

MAXGRN specifies the maximum number ofgranules assigned to a file when no maximumis provided.

89

FACILITIES PARAMETER (Cont'd)

MAXSEC SW, FALL, 83, EQU 12 5

MAXSEC specifies the maximum number of massstorage sectors containing file granuletables that may be requested at one time.FALL checks if the number of requestedsectors is greater than MAXSEC. If therequested number is greater than MAXSEC,MAXSEC is used instead. MAXSEC is used inthe same manner to determine the requiredEXPOOL request buffer size for the elementFASEC. FASEC is the EXEC element thatallocates sectors of mass storage space forEXEC use.

MSWl CP, AACONFIG, 1012, EQU MSW2+MSW4

MSW2 CP, AACONFIG, 1011, EQU MSW3+MSW4

MSW3 CP, AACONFIG, 1009 , EQU 0150*MAXOPN

MSW4 CP, AACONFIG, 1010 , EQU MSW3/2

MSW5 CP, AACONFIG, 1013 , EQU 0200

MSTOL CP, AACONFIG, 1014 , EQU MSW4/4

MSWl through MSW5 are thresholds used in therollout process to determine the criticalityof mass storage availability. A consolemessage is displayed whenever mass storageavailability goes below each threshold, anda different message is displayed when theavailability exceeds each threshold. MSTOLmakes the amount of available mass storageappear less than it actually is when massstorage availability increases. Thus, MSTOLprevents "toggling" between console messageswhen mass storage oscillates around a thresh-old.

NGTB SW, FALL, 82, EQU 5

NGTB specifies the maximum number of granuletables that a catalogued file may have in

90

FACILITIES PARAMETERS (Cont'd)

EXPOOL when EXPOOL becomes 'tight'. EXPOOLis 'tight' when the available EXPOOL Storagedrops below EXTLOW words (an EXPOOL parameter)threshold. When a low-priority EXPOOLrequest is refused during the addition ofanother granule table, the number of granuletables in EXPOOL is reduced to NGTB.

PERCNT SW, ROLOUT, 11, EQU 10

PERCNT specifies the fraction of totalsystem mass storage tracks to be made avail-able as a result of a rollout. A rollout isinitiated when the number of available(unassigned) mass storage tracks drops belowthe threshold MSWl . If the amount of massstorage available at the time of the rolloutis denoted by AV, the number of tracks(NBTRK) to be unloaded to tape is given by

NBTRK = TCPERCNT

where TC is the total mass storage capacityin tracks configured on the system.

NBTRK is constrained so that it will be inthe interval

[MSWl, 3OOOO0]o

That is, at least MSWl tracks, but no morethan 30,000p tracks, will be unloaded foreach rollout.

POSDIR CP, AACONFIG, 1099, EQU 191

POS60 CP, AACONFIG, 1107, EQU 135

POSDIR specifies the position of the firstdirectory track on a FASTRAND II. It alsospecifies the position of a FASTRAND Ill'sfirst directory track. For a FASTRAND III,POSDIR*3/2 is used. POS60 specifies the

91

FACILITIES PARAMETERS ( Cont ' d

)

SIZEWT

SSDENS

position of the first directory track on aF8460. For all other devices, the firstdirectory track is located in position 0.

SW, SECURE (MAIN), 738, +1

SIZEWT is the weight assigned to file sizein tracks in the computation of the unloadeligibility factor (UEF) . This settingapplies to LEVEL 19R1 VERS 2 of SECURE.

CP, AACONFIG, 611, EQU 'V

SSDENS specifies the density in flux changesper inch (FPI) of tapes assigned by thesystem. The codes for SSDENS are the sameas those used on the ^ASG control card. Inthis case, a density of 1600 FPI is beingused

.

SSEQPT

SYSPRP

CP, AACONFIG, 614, EQU '16N'

SSFQPT specifies the equipment type alltapes assigned by the system will use. Inthis case, a UNISERVO 16 nine-track tape isspecified

.

CP, AACONFIG, 621, EQU 4

TFEXP

TFMAX

SYSPRP defines the system prep factor forthe EXEC resident on the system disk unit.SYSPRP is the number of FASTRAND (28 word)sectors that will constitute one disk record.SYSPRP should be set to either 1 or 4 ifdisks are configured. This will give either28 or 112 words per record.

CP, AACONFIG, 633, EQU 6 0

CP, AACONFIG, 1327, EQU 360

TFEXP is a systems standard retention periodfor tape files in days. TFMAX is the maximumretention period for tape files in days.Both TFEXP and TFMAX are limited to a maximumspecified value of 2047 because they arestored in 12-bit fields.

92

FACILITIES MANAGEMENT (Cont'd)

SW, SECURE (SETUP), 2425, See V.B.4

UEFLUP is a look-up table for obtaining(1) the point values assigned to timesince last reference (TSLR) and (2) themean time between references (MTBR) inthe unload eligibility factor (UEF)computation

.

SW, SECURE (SETUP), 2430, See V.B.4

UEFTRK is a look-up table for obtainingthe point values assigned to file sizein tracks. It is used in the computationof the unload eligibility factor (UEF)

.

CP, AACONFIG, 65 9, EQU 1

When WHLDRM is set to 1, word addressable,unit-granular assignments will be made.WHLDRM is a conditional assembly flag thatturns on code to handle unit-granularassignments

.

SCHEDULER PARAMETERS

APRIOR CP

,

AACONFIG, 285, EQU • s •

CDHOLD SW, LOCTAB, 53, +20

CHGDED SW, LOCTAB, 2, EQU 3

DEDLIN CP

,

AACONFIG, 363, EQU 1

MAXCRD CP

,

AACONFIG

,

506 , EQU 500

MAXCRQ SW, LOCTAB , 1

,

EQU 2

MAXDEM SW, AAPCT, 1419, EQU RSICNT

MAXOPN CP , AACONFIG, 522 , EQU 8

MAXPAG CP

,

AACONFIG, 527 , EQU 300

MAXTIM CP , AACONFIG, 531

,

EQU 10

MAXTUX CP

,

AACONFIG, 536 , EQU 30

MPRIOR CP

,

AACONFIG, 540 , EQU 'A'

NHOLDT CP , AACONFIG, 1016 , EQU 16

RSICNT CP, AACONFIG, 595, EQU 80

TRMXCO CP, AACONFIG, 642 , EQU 0

TRMXPO CP, AACONFIG, 646 , EQU 1

TRMXT CP, AACONFIG, 650 , EQU 1

94

SCHEDULER PARAMETERS

CP, AACONFIG, 28 5, EQU 'S'

APRIOR is the default priority for (SRUN

statements that have no priority specified.APRIOR may be any alphabetic characterselected from the set A through Z. Thenearer APRIOR is to the beginning of thealphabet, the higher the default prioritywill be.

SW, LOCTAB, 53, +20

CDHOLD specifies the maximum number ofactive demand runs. When the @RUNstatement is received from a terminal,it is determined whether this additionaldemand run will cause the number ofactive demand runs to be greater thanCDHOLD. If the number of active demandruns will be greater than CDHOLD, (1)

the user is informed that a system holdexists on the number of demand runs and(2) the information about the run isreleased.

SW, LOCTAB, 2, EQU 3

CHGDED specifies in minutes when adeadline run's scheduling priority ischanged to critical as the run's dead-line time approaches. The run's priorityis changed to one of the priorities 2

through 5. The priority is changed to'n' when the run's deadline time isCHGDED*n minutes away, where n can equal2, 3, 4 or 5.

CP, AACONFIG, 36 3, EQU 1

Allow deadline runs if DEDLIN is non-zero .

CP, AACONFIG, 506, EQU 500

MAXCRD is the default for the maximumnumber of cards punched for batch runs,if not specified in the ORUN statement.

95

I

SCHEDULER PARAMETERS (Cont'd)

For a demand run without a maximum number ofpunched cards specified, 131071 will beused. Whether the maximum is specified ornot, it is limited to 262143.

MAXCRQ SW, LOCTAB, 1, EQU 2

MAXCRQ specifies the minimum number of corequeue entries that the Coarse Schedulershould provide to the Dynamic Allocator. Itis defined, but apparently never used.

MAXDEM SW, AAPCT, 1419, EQU RSICNT

MAXDEM is a conditional assembly parameterfor turning on demand code when MAXDEM isgreater than 0. MAXDEM is also used indetermining the swap file size. RSICNT isdescribed below.

MAXOPN CP, AACONFIG, 52 2, EQU 8

MAXOPN is the maximum number of batch runsthat may be open at one time.

MAXPAG CP, AACONFIG, 527, EQU 300

If the estimated page count is unspecifiedin the QRUN statement, the default for batchruns is MAXPAG. The, value used will belimited to 262143 (2-1) . If the P-optionon the @RUN card is specified, a run isterminated when its page count is exceeded.Otherwise, the operator is notified andgiven the option of terminating the run.

MAXTIM CP, AACONFIG, 531, EQU 10

If the estimated run time is unspecified inthe (9RUN statement, the default for batchruns is I^IAXTIM in minutes. If the T-optionon the @RUN card is specified, a run isterminated when its estimated run time isexceeded. Otherwise, the operator is notifiedand given the option of terminating the run.

96

SCHEDULER PARAMETERS (Cont'd)

CP, AACONFIG, 536, EQU 30

MAXTUX is the number of seconds beforereaching the estimated run time at which theMAXTIM contingency will be honored.

CP, AACONFIG, 54 0, EQU 'A'

MPRIOR is the maximum allowed priority on a(3RUN statement. If the priority specifiedon the @RUN card is higher than MPRIOR,MPRIOR is substituted. MPRIOR may be anyalphabetic character selected from the set Athrough Z. The nearer MPRIOR is to thebeginning of the alphabet, the higher themaximum allowed priority will be.

CP, AACONFIG, 1016, EQU 16

NHOLDT is the maximum number of attempts toopen a run from the Hold Queue before theoperator is notified. Runs are placed inthe Hold Queue if facilities are unavailable.NHOLDT must be less than or equal to 077.

CP, AACONFIG, 59 5, EQU 8 0

RSICNT is the maximum number of terminalsallowed to interface simultaneously with theRemote Symbiont Interface (RSI) . When aterminal signs on, a check is made todetermine if the number of active terminalsis greater than RSICNT. If the number ofactive terminals is greater, the line isterminated. RSICNT is also used as a con-ditional assembly flag to turn on demandcode when RSICNT is greater than 0.

CP, AACONFIG, 64 2, EQU 0

If non-zero, TRMXCO terminates a run whenthe punched card estimate is exceeded.TRMXCO has no effect on demand runs. Itoverrules the "C" option on the (SRUN card.

*

TRMXPO

TRMXT

SCHEDULER PARAMETERS (Cont'd)

CP,, AACONFIG, 64 6 , EQU 1

If non-zero, TRMXPO terminates a run whenthe page estimate is exceeded. This has noeffect on demand runs. It overrules the 'P'

option on @RUN card.

CP, AACONFIG, 650, EQU 1

If non-zero, TRMXT terminates a run whenthe estimated run time is exceeded. Thishas no effect on demand runs. It overrulesthe 'T' option on @RUN card.

93

I/O PARAMETERS

ANGLOF

DHLACE

DHLACH

DHLAC8

FASTSK

INCORE

lOLM

MAXF4

MAXF8

MAXF17

MAXLLC

MAXOPR

NOSEEK

POOLSIZE

SCGA

SYMALL

SW,

CP

CP

CP

CP

CP

CP

CP

CP

CP

SW,

SW,

SW,

CP

CP

SW,

AAPCT, 93 3, EQU 1

AACONFIG, 66 8, EQU 0

AACONFIG

AACONFIG

AACONFIG

AACONFIG

AACONFIG

AACONFIG

AACONFIG

AACONFIG

10, 3561

10, 3560

AACONFIG

AACONFIG

AASMTAGS

671, EQU 1

674, EQU 0

379, EQU 0

416, EQU 1

451, EQU 1

510, EQU 200

513, EQU 10000

516, EQU 1000

EQU 30

EQU 12

10, 80, EQU 1-STPAUL

1093, EQU 25

688, EQU 0

3598, EQU 036

99

I/OPARAMETERS

ANGLOF SW, AAPCT, 93 3, EQU 1

ANGLOF deteaddressingtional posidisk subsysconditionalangular addwise, angulANGLOF is s

(FCO) numbe5033 Controuse angular

rmines whether the angularcapability (also known as rota-tion sensing, RPS) on the 8440tem will be used. ANGLOF is aassembly parameter that enables

ressing if equated to 0. Other-ar addressing is disabled whenet to 1. Field Change Orderr 037 must be implemented on the1 Unit before the 8440 disks mayaddressing.

DHLACE

DHLACH

DHLAC8

CP, AACONFIG, 6 68, EQU 0

CP, AACONFIG, 671, EQU 1

CP, AACONFIG, 674, EQU 0

DHLACE, DHLACH, and DHLAC8 are the interlacefactors for 432, 1782, and 880 drums. Thepossible values are 1, 2, 4, 8 and 16; thesevalues should correspond to the interlacefactor at which the drum is actually set.An interlace of 1 implies that consecutivedata words are stored in consecutive drumlocations. If the drum interlace is n, n-1drum locations are skipped between consecu-tive data words. Doubling the interlacefactor halves the drum transfer rate.

FASTSK

INCORE

CP, AACONFIG, 379, EQU 0

If FASTSK equals 1, all subsequently preppedpacks will have their last simulated FASTRANDposition marked as allocated. That theinner cylinder remains unused results in areduction of full pack seeks.

CP, AACONFIG, 416, EQU 1

INCORE specifies the method of processing ascatter read or gather write request on adisk subsystem. If INCORE equals 1, the

100

I/O PARAMETERS (Cont ' d)

EXEC insures that access control words(ACW's) fall on record boundaries. If ACW

s

do not all fall on record boundaries, thecode generated by INCORE will build a newset of ACW's that do fall on record bound-aries .

CP, AACONFIG, 4 51, EQU 1

lOLM is a conditional assembly parameterthat controls the I/O Load Monitor codegeneration. I/O Load Monitor adjusts theI/O load on memory modules that may beoverloaded by I/O requests. If a memorymodule's maximum I/O transfer rate is lessthan the combined transfer rate of thesubsystems active on the memory module , alate acknowledgement (data overrun) canoccur

.

CP, AACONFIG, 510, EQU 2 00

CP, AACONFIG, 513, EQU 10000

CP, AACONFIG, 516, EQU 1000

MAXF17 specifies the number of FH-432 (MAXF4)

,

FH-880 (MAXF8) , and FH-1782 (MAXF17) tracksavailable to the symbiont complex for writing.This FASTRAND-formatted mass storage allocationfor user symbiont files is accomplished by a

quota system based on the number of usersymbiont files. A file will obtain a certaintrack allocation quota on each type of drum.The quotas will be used by a file in theorder (1) FH-432, (2) FH-1782, and (3) FH-880. If the quotas on these three drums arealready used, the remaining amount of theuser symbiont file will be allocated on thedevice specified by the parameter SYMALL.

SW, 10, 3561, EQU 30

I4AXLLC is the maximum number of times a

queued I/O request may be locked out by on-position requests.

101

I/O PARAMETERS (Cont'd)

MAXOPR

NOSEEK

SW, 10, 3560, EQU 12

MAXOPR is the maximum number of consecutiveI/O requests that will be satisfied for adisk unit's current position before theposition will be forcibly changed.

SW, 10, 80, EQU 1-STPAUL

NOSEEK is a conditiothat enables 8440 diNOSEEK equated to 1

pre-seeking off and(STPAUL = 1) enablesField Change Order (

implemented on the d8440 pre-seeking isNOSEEK to 0.

nal assembly parametersk unit pre-seeking.(STPAUL = 0) turns 8440NOSEEK equated to 0

8440 pre-seeking.FCO) Number 037 must beisk controller beforeenabled by setting

POOLSIZE CP, AACONFIG, 1093, EQU 25

POOLSIZE spqueue itemwithin therepresentsIf POOLSIZEwill occurinstead. Uthe elementbuffers wil

ecifies the number of 15-word I/Obuffers (IQ items) maintainedEXEC element 10. Each IQ iteman I/O request that is queued.equals 0, no pooling of IQ items

and EXPOOL buffers will be usedse of I/O queue buffers within10 reduces EXPOOL overhead, but

1 be assigned to EXPOOL if required,

SCGA

SYMALL

CP, AACONFIG, 68 8, EQU 0

SCGA specifies whether scatter read andgather write I/O will be used by the EXECsymbiont element SYMBIO. SCGA greater thanzero will enable the use of scatter/gatherI/O by SYMBIO. No scatter/ gather I/O willbe done if SCGA equals 0.

SW, AASMTAGS, 3 598, EQU 03 6

SYMALL specifies the equipment code of themass storage device used for user symbiontfile allocation after the file's quotas on

102

I/O PARAMETER (Cont'd)

FH-432, FH-1782, and FH-880 are used. Thefile's quotas for FH-432, FH-1782, and FH-880 are specified by the parameters MAXF4,MAXF17, and MAXF8, respectively. In thiscase, 036 specifies that the FASTRAND-formatted 8440 disk subsystem will be used.

103

TEST AND DEBUG PARAMETERS

DADBUG

DATRAC

EXPTRACE

FCDBSZ

FNCCHK

lODBUG

SBLEVl

SBLEV2

SBLEV3

SBLEV4

SBLEV5

SIPDAT

SIPIO

STPAUL

SWPCKO

UN IVAC

SW

SW

CP

CP

CP

CP

CP

CP

CP

CP

CP

CP

CP

CP

SW

CP

AADAEQT, 3, EQU STPAUL

AADAEQT, 9, EQU 30*UNIVAC

AACONFIG, 896, EQU 130

AACONFIG, 922, EQU 50

AACONFIG, 401, EQU 0

AACONFIG, 97 2, EQU 1

AACONFIG, 1199, EQU 1

AACONFIG, 1206, EQU 1

AACONFIG, 1210, EQU 1

AACONFIG, 1214, EQU 1

AACONFIG, 1218, EQU 1

AACONFIG, 1222, EQU 1

AACONFIG, 1229, EQU 1

AACONFIG, 2 80, EQU 0

AADAEQT, 5, EQU UNIVAC+2 *STPAUL

AACONFIG, 654, EQU 1

104

TEST AND DEBUG PARAMETERS

DADBUG

DATRAC

EXPTRACE

FCDBSZ

FNCCHK

lODBUG

SW, AADAEQT, 3, EQU STPAUL

DADBUG specifies whether additional DynamicAllocator internal error detection is done.When DADBUG is equated to 1, error detectionoccurs

.

SW, AADAEQT, 9, EQU 3 0*UNIVAC

DATRAC specifies the size of the swaplock tracearea. It must be an even number whose value is2 -2, where n is any positive integer.

CP, AACONFIG, 896, EQU 13 0

EXPTRACE is the number of EXPOOL requests andrelease trace entries. Each entry consists oftwo words

.

CP, AACONFIG, 922, EQU 130

FCDBSZ specifies the number of FNCCNT actionstraced. Each action is traced by a five-wordtrace entry.

CP, AACONFIG, 4 01, EQU 0

FNCCHK determines whether function validationis performed whenever a nonresident function isread from drum. If FNCCHK equals 0, nochecking is done. The function's keyword ischecked when FNCCHK is 1; a checksum iscomputed when FNCCHK equals 2.

CP, AACONFIG, 972, EQU 1

SBLEVl

SBLEV2

SBLEV3

SBLEV4

When lODBUG equals 1, all debug aids withinthe I/O complex are enabled.

CP, AACONFIG, 1199, EQU 1

CP, AACONFIG, 1206, EQU 1

CP, AACONFIG, 1210, EQU 1

CP, AACONFIG, 1214, EQU 1

105

TEST AND DEBUG PARAMETER (Cont'd)

SBLEV5 CP, AACONFIG, 1218, EQU 1

SIPDAT

These parameters turn on and off the variouslevels of SIP. The SIP levels are discussedin detail in the appendix on SIP.

CP, AACONFIG, 1222, EQU 1

The SIP Dynamic Allocator Trace is enabledand written to the SIP SYSBAL$LOG$ file whenSIPDAT equals 1.

SIPIO

STPAUL

CP, AACONFIG, 1229, EQU 1

The SIP I/O Trace is enabled when SIPIO isequated to 1.

CP, AACONFIG, 280, EQU 0

Extensive internal EXEC error checking isdone. STPAUL is set to 0 for all sitesthat are not STPAUL sites.

SWPCKO SW, AADAEQT, 5, EQU UNIVAC+2 *STPAUL

When SWPCKO is non-zero, program bank swapchecksums are calculated. Two words perblock are summed if SWPCKO equals 1; and,sixteen words per block are summed if SWPCKOis 2. If SWPCKO is 3 or more, every wordin the bank is used in calculating thechecksum.

UN IVAC CP, AACONFIG, 654, EQU 1

When UNIVAC is non-zero, the primary EXECerror checking is enabled.

106

APPENDIX C

INTRODUCTION TO SIP

107

SIP is an event-driven software monitor. That is, SIP

counts and times the occurrence of significant events withinthe operating system. The data collection code is integr-ated into the operating system. Special instructions areincluded in the executive programs to count the frequency ofcertain events or to time the duration of certain activities.These instructions are preceded by conditional assemblyflags. The conditional assembly flags are turned on or offby system generation parameters. When the flags are off,the SIP data collection code is not generated as part of theoperating system. Therefore, if SIP is not going to beused, the SIP overhead may be avoided.

SIP data collection is hierarchically defined. Fivelevels of detail may be specified at system generation time.The lowest level generates the least overhead and collectsonly a basic set of data. Each higher level of SIP includesan additional degree of detail. The levels are divided intotwo classes: user levels and development levels. Userlevels collect basic data of general interest; developmentlevels collect detailed information needed for operatingsystem tuning. There are two user levels: User-A and User-B;and three development levels: Development-A, Development-B

,

and Development-C. Each level has a separate conditionalassembly parameter associated with it. These parametersdetermine which SIP code is or is not assembled when theoperating system is generated. The data collected by eachlevel are summarized in Table C-1. Each item in Table C-1corresponds to a standard report.

Three programs support the SIP data collection code:SYSBAL, SYSSIP, and SIPDRP. SYSBAL contains the storageareas needed to hold the data collected by SIP. SYSBAL alsocontains numerous subroutines that support the data collec-tion process. SYSBAL must be permanently core-resident whenSIP is on. SYSSIP is a nonresident element that providesthe interface among the user, the operator, and the system.SYSSIP is capable of responding to a large number of operatorkey-ins. SIPDRP is the SIP Data Reduction Program; it sum-marizes the SIP data and produces a set of standard reports.SIPDRP is a FORTRAN program that may be run whenever it isconvenient

.

The operator key-ins provide a substantial amount offlexibility in controlling the operation of SIP. Of particularinterest is the SB CODES S1S2S3S4S5S6 key-in (all SIPoperator key-ins begin with SB). The codes' key-in may be

109

SIP REPORTSIP LEVEIS

1 2 3 4 5

System Balance Sunmajry 1rtonory Utilization 2

Exec Activity for Each CPU 2

Dist. of Prog. Size for Run Type 2

Processor Activity for Each CPU 1

Idle Tine by SIP Collection 2

I/O Activity for Each CPU 1

Mass Storage SS Queuing by Type 1Mass Storage Analysis by Unit 1I/O Queuing by SS 1

,

Pre-Seeking and Pre-Positioning 1

SWSP$FILE Expansions 1Facilities Hold Queue Statistics 1

I/O Load Monitor Activity 1

Common Bank Usage 2

CTMC Comm. Line Usage 1

EXPOOL Requests and Releases 3

EXPOOL Counts by Priority 4

Times EXPOOL Became Tight 1

EXPOOL Usage 1

Register Saves and Restores . 1

Segment Requests and Loads 3

Segment Initial/Active Requests 4

System Idle Activity 1

ER Counts 1

Response Times (Abbreviated) 2

Response Times (Complete) 3

Core Fragmentation 3

Suspend/Swap Statistics 3

Exec Frequency Counts 3

Test and Set Summary 5

SIP Overhead 1

SIP REPORTS (LEVEL 4.0)

TABLE C-1

110

used selectively to turn on (SI = 1) or off (SI = 0) six areasof SIP data collection:

SI - User level A data

S2 - Distribution of user program sizes

S3 - Memory-in-use distributions

S4 - EXPOOL statistics

S5 - Function/Segment Activity

S6 - Most of the rest

The status of each of these codes is reported on the firstpage of the SIP output. Within the operating system, theSIP data collection code is still present; however, if theappropriate status code is set, a branch instruction by-passes the SIP instructions.

The status codes are very helpful in reducing the amountof overhead produced by SIP. Since SIP is event-driven, SIPoverhead is proportional to the overall activity of thesystem. The overhead generated by SIP is displayed on thelast page of the SIP output. The overhead associated withprogram size calculations is also reported. Since programsize calculations often account for 50% or more of the totalSIP overhead, it is usually a good idea to turn off the S2status code. Another way to minimize the amount of SIPoverhead is to run SIP periodically for short periodsrather than continuously. Experience has shown that 10- or15-minute periods are long enough to observe steady statebehavior. The starting and stopping of SIP are easily con-trolled with operator key-ins.

The remainder of this appendix is a set of annotated SIPreports. Special comments have been typed on these reportsto aid in their interpretation. The reports are in the samesequence that they were printed by SIP. Detailed informa-tion about SIP may be found in the SIP documentation manual.

Ill

<

ISDCO

u«<

•J«<

PC

:g

H

in

I

IVti

<M

CVI

n

00

OMX(U

0.H

CO

<

10

Unso

X)•J

n

CO

H

H

P

<

W><

Hz

to

o

•J

•J•<!

(C(0!h

CO

0^

H

CO

nH

De;

ID

W

Wn

TD

H

<!CD

CO

>^

CO

HDa.

HD«•

(b

ID

to

MO

m

U)

•J

hID

CO

<

n

P

<>-Qd•<

&CO

wH

>MHy

CO

p

CO

p

outo

CO

toH

w f •

in

•

to nTI ZI GE IC

toST

in If)

fiiwHM I* Ui WCVi > 1—'

iT) (L •< H w WCM Ft

)

rjw (Li

CLi

to wi © ID CO B5

til rv U ID

CO .» TP.

CU oo <!

PC <; Lj M in

W M-lt . f\(M w/

nw LU M >^TPi •

><! H to p H w 5Sm

win• t

(L. p to 00TP o

Ui

p oID z z Z z z Z o

M U ID ID ID in

•< CO H tt

P 1 1 1 1 1 1

< M zCO t-1 (VJ ro <t in \0 H

CO CO to CO CO to HW w ZK K PH H

as

55 OiM

TI LA

to WM

PU <M HCO ID

H

(0

MCO

n

112

tit

0>

><

KO0*

un

wnH

M Z>)J Ck!

M fU

HD 2U

in WW CO

W CO

MHp

oUJX

>1

5

i

U I

Ui I

W I

w I

I

I

I

ooooooooooooo^o\cr) toooooooooooo»nooN<f i isiOOOOOOOOOOOON<OvOOO I o>

O 00 I 0^

h- in I rn

rn n I in

ooooooooooo o f ro 00 r< I n(vj a C\J I •*

fO 00 I wI ^

0000000000000(\JW>t)

US 1^

in 00

ro 00 0^

w tn CO CO CO CO CO CO CO CO CO CO CO CO 1

Ui M Ui M Ui •

o O o o u o V u o u V 1 I-I

ID ID 1 •<

hJ iJ iJ I-I h4 1 HCQ pa m CQ CD CQ CC ffl cn CD CD CP CQ CD 03 CQ I ID

H(VJ * o oo o W * >£) 00 o CM •* 00 o (\J

1

1

n (Vi 0> CM in 00 (M CO * oo rH 1

fl-t T-1 CM CM CM n f) * «

W(M w

CO

w

[II

4*

HM

n ^

< (0j; NID M (14

,J oCO It, wP M

H«(; CO

M tsl

M M>< w ,J

pel o M >4

Z; CD

<w K

CO >C

6S wc

'O •g <;

jEM

Cl r)wCO (-1 W

COrjwJr Q M«CD IP w f ^w

Qi QCV C

< &o o u.

o ID zH C M

S- R ><

1 CK

ID CM -<

a MH 5! in z Hz <

CO H pID Cfi

o oct; C^

>- M u <1^ z >z T M

•< D P z PH p CO

*ff wtt CO CO

tt u: phJ u oHi Ui

»- •• >H CD

113

O -r-

4—* 1'

1 1] QJ

(_) "—1 E

1 ' cTi ri ^

•r—

QJE

. ? 5-t->

O M—O fD O(/I fD

LU E OX OJ LOLU on ct -OQ r—+->

LO LU OJ =JO X -c: c CTLU 4-) -1— OJ

oooooooooooo

Hzwc.

H o o o o o10

u

o o o o o o oo

o <t IDo CO

-* <*

w •

s in1-4 CD

H

w

>< o mH u u u

w u uu X X Xw wX[11

•<

Cki

OID

oe

wwp

t

H

nH0!

Hw

ooooooooooooooooooooooooooooooooo

oooooooooo O N

OOvO'4(\JO(J>f~-irOP»<>0>0

OOOOOOOOOOO^OOOOOOOOOOOOOOOOOOOOOOOO

OOOOOOOOOOOO"

<ci£'-<t(\ioa'r-incD»Hif)i£)

114

'v) "O (L) 1

—

C71 O, q_ •1— >

^ -{B 4-J QJ> O C 1— -M

(fl

M

WQJ -Cra a> +-> (/I t/1 +-> s

Ma> S- C ^ C Hf~ +-> O) 3 ••-

+-> x: c ai+J _c:

C 4-> (D 03 fO 1/1 -i-> >4I/) to -CJ -!-> -C E - <

5 > -,- c +-> ro 3 H-i-> s- in <U s-

0) c w QJ (_) OJ C7' t/)<C

E O) OJ s- M o a; HCL-i- 0) T- s- •.-

(/) 1

—

&^ Q. I/) O-i—

o Q

(D o> 0^ o> (J> 0^ o>

ooooooooooooo

w<tooooooooo

to t o (;> in O o o o o o o ID o o o o o o o o o vO vO in o o o o o o o o oin o 00 ^ O o o o o o o (vr o o o o o o o o o o o o o o o o o o o oro o n N o o o o o

MEo o o o o o o o o o o o

ME10 o ^ o o o o o o o o o

w(VJ H TI

00

iDi^ in»<r>if)ooooooo iH If) o> m o (VI

(vi .H 4 ro CO <vi in<vi «t

ooinooooooooo ro -I oD

vt) n in

« ^

ooooooooo

co<o*(vioo>r^in<i)iH\o\D«<vjro*<fin>oo>r)(yov

+->

oQJU to ^ s_X ^ c QJ QJ

E QJ to X2 EfO 1— ^ E tos_ o i/i

CD (/I D- QJ to CDo 3 o tos- O -!-> QJ S- >D- to c c O- t/) s-

to +-> +-• o QJ QJn3 o tO +->

1/1 c l/l +-> CM- +J oO to c s- QJ u +J QJ

to o a QJ M X)QJ 1/1 O X -o QJN OJ s- lA u QJ QJ I/)a C QJ QJ sz H-(/) 3 O ^ S- >-, in o QJ

E +J QJ +-> 1/1 UQJ U E QJ 0) E O QJ^ c O x: S- to S x: Q.1— u 3 lO I— +-> i/i 3 1/1

a)<0'*<vioo>h-ino)»<\Ovo»Hcvii«)it<*initJ(;>ioo>o^

M M M M M tx: UlvHtf M ut

oo>0'*(vioo>h-inco»<voio•-•{vin-*<tiniot7ina'Oi

115

uato

\H

>>

•!->

s_ -o0) +J QJ o> -a CL0) .— -r- O)

co > !-

E !- 3 -r-

3 crT3 S_+-> O OJ

O l/lte

o O "O TDI— Co l/l 3 O

UJ •4-> O O (OX Q-x: 0)LjJ 3 (/) (/I

s-c:

i- <U Q-+J QJ 3 1—

1

(->

4-> 1— 00 3s c to

•1- > -oo

cu <U c4-> ^ (/» tn cu(0 O -1- Q-

O -C (D o'ai 1— +-> 1

—

s_s_ U <D J3S-o >, l/l M- Olo (O "O O s-

03a O i.Ol O OJ OJO <U (/I

o C c/l E OJ_c •1- 3 _Coo OO V£3 C l£) 1—

Oir(MO<VJin(DOOO<M<*.HOOOOO>DO<\J If) *

enS-

E3C<U

<u_c1—

1/1 1—UJ

>>JD =c

-o cOJ oS_3 o(/I OJfOOJ fT3

ECO tu

<o EQJ -!->

EOJ

4-> r--a "O

OJ

-o XI oOJ+->

m X)TO E3 3+-> +-> O

UJ

T7

w p^ 0.

M K

M >- o to to O CVI in o o o 0' in in t-i o o o o * o '0 nH hJ If) w o m o w 0)

U >J «£) O in oo s Q a•< «: 0- (M (VJ w M Z

(VI Hu U

c 0^ (i. wU3 w wm pi

M in N <c *U O in ^ in

X> ^ (0 to w inOS in w •

Ou •* (D

HO

a<

t)c>j

wi-i CO

Q PM Z

X)D ON U

U <- (0

o ^Oo•

in

<to

0> o WinCVJ

IDL

•

< CO u\ ID H < zw i-i U. HI OT M

&> H

Z w M H TO05 « H OS T> o (0 M <J H U H <

K OS 06 M > z H H Q O H» w H Z H <D © p U M M < P. M HI

N « H <c H H H H M U B) M u w TO W >H w M z Z »H z P H H (- W u TO HI MZ u z Z TO Z K Z z o P z < < hJ K HM d x> X I-I kI w H w H H z z w p H H H

Bi •< iZ < u (fl M TO hI < < M z z W Z <U. Hi M z H z w Z w 0. a> P w uc K H p H K H P H P ce w u M p < OS z w tu u D H

W P 0. U W P Ph P a lU o c O O |i< in to TO z rs PS OS HI Pu ^1 Ol H H 0. H QU H H OS l-< z z z M M l/J M HI w m •< a.

z z P P z P Z P X J H B. < H M IK l-< l-i VJ Q a. 01 z eM lb M M <D H <C M » p H P w W >< » p

H i-l a w •< < IH y K u UJ lu w w m TO TO•< i- •«! TO TO > w W TO TO >J hJ h) hI Z

V) w (Q w to in w w M «; M z P i-l >-l M Vl OS TO Q P p p n Hu M m PtS M M o U. Q P 0. >-l HI M < OS

116

©

d>

wo q

o <

•z.

MHU•<

PCu

1 If If IF If If If iF IF If ir IF IF If IF fl IF If If IF IF Ir Ir ?F If If If If if IP

1

1

ro O O O O O o O (/ O o O O 1

1

ro O O O O O i-« o o o o o

1 IP IP 1 *P tP tP iP IP IP IP *P IP IP tp tp iP

1j

ro

1

as

PhtP tP tP tP 1P iP tP IP tP iP Ip tp IP tP (1.

j• DO o o o 00 o Q 1 CO o Q o o Q CO o ro Q o X*^ *^ CM

cH

1 If Ir If_If (F

_ _If

_If

_ _If )* IF »*

-If IF

_If

_If

_ -IP

-IF

_ - - ED

1 CJ OJ O o 1 <\j O O ro o1 r- 0^ (7* 1 in (D M

•

o wU zPS

ww

"!

M1

- D .inUJ m

j

m _

In

_ m m _ Hj

OJ in

I

ni in fti

(-1

j I tc

j: H

1:

1

1 1 ©Z H

to P_ n _ m m (VI— — Q - Q ~

ao_

[uj 00 r) OD OS in CM1 in Q\

!OD in lO m (M < M

* cu (M

CO

,J

wz

w z wz <

HI ^ ITOJ in *^ CM z

1 UJ CD UJ 1 tn in CM ro AlCM H1 If)

Jn j

(M

5

*"* D OS W1 1 w P

(1

s2:Hz

M OLi Mp

I lO Ct\. —

in CM w H* m

JrinU) in

Si u5

r Uj m1

rn 0" Al CD Z1 o in

1fn

tn At 1^ win \0 oo CD JL 1^ rvi TO m

1 O U) 1 »o pq Mj

*^5

r) u1 M P

i 1 D EC z

H WP ^

Z_P

1 OD— _

UJ— _ — ^

trn If 1

1 1

' If) tn_ —

1 in_ _ — m - — — _

1in"J

in

In In1 (7* in o

vlin

SiID Z

1 m u) XO O 'O o 1 o o O in vn z lb

1 UJ 1 M1 rj

In* O 10

jD CM hJ CD m

** X

HID

H u HT> >

»-(

H uU

(\J n EQ PC <VJ PC Eu CM «: u*H u ? Q O U u ? M w b«

^ W w (VJ M w Al

CT\—

ww EC r] 1^ z

w M H N N hj N H M w W Eb p M lO N r M w z z<B

O u Pu

1 O W r> ^ in ^ 00 O <VI ro in 1 O •H CM ro in OO o r4 CM ro in (U w• 1 aa tc pa

1 H t H H1 I-) •-I •J 1-1 i-i P 1 hJ i-l .J *J IJ 1-1 1-1 i-l Hi

1 w U u w m m m u Ou 1 w U u u m m Ul w Hi

1 2; z z z Z z z z z z z z Z z z z t- 1 Z z z z Z z z Z z z z z z z z Z1 z z z z z z z z z z z z z z z z P 1 Z z z z z z z z z z z z z z z z U1 < •< -< < < < < < < •< < < < < < < 1 •< < < < < < -< < < •< < <; < < •< H1 Ed RO m EC n V m tn to K 1 K PC pa EC IE CO pa tr s PC PC « PC tc PC

1 o o u u u u V U u B U u u u u y V u V o o V Z

3 3

o o

fT3

"O "O 4-> Ol(U 3 Q- C!- 1— OJ C

S- OJ QJ X- OJ CD ufD D-

Q.1—. _ ,. _Oji/OM-cci/iOJ-c:

+-> C S~ S- C O" n3Ol

+->+-> OU OJ

o .— <u3 E -O

•,- •.- OlJ M- E +->

Oi- •!- l/l E

: OJ 3 O> -Q O ^: E- C o i-

I— OJfT3 UO OJI— O-

I— Ol

Ei- 01 U Q-<D E 1- !->

> 01 3<c >— a. a

ct CO (_> Q 1

117

o o

t-l

Hp

iJ 0.•< oH IK

^ o <0 PS

tt.

p cn

wN

OT

3<KOM n

> o o *M o in o hH • • •

U s rH BS

< 0> n UH (C

fu p>t-t d.HDy z

X Mm

W CQa, H

o ro wu Uw w w to

w X X X HI

w w u

HOW ooooooooooooCO

ooooooooooo*ooooooooooowOOOOOOOOOOOCVI

ooooooooooo

a3<0*(VIO(7>NlflOO»4i£)<0

oooooooooooo

OOOOOOOOOOO!"ooooooooooooOOOOOOOOOOOO

OOOOOOOOO

(t)>0^(\JOaNI/)(D>4<0>0

P 8

118

o

(i) Q\ Q\ Q\ 0^ 0^

ooooooooo

o o o N w <t o o o o o o o o o

in o o 00 o o o o o o o <0 o o o o o o o o o <£ flH O o o o o o o o om 00 o o o o o o o r> o o o o o o o o o o ^ o o o o o o o o o<t ro <c in <0 o o o o o u o o o o o o o o o o o o \n o 03 o o o o o o o o o

\C n iH

<l-i£IWW<l-lOOOOOOO>0 (M h <\J O IDo iH 1^ r) w <<vi n

o o ooooooooo

HD

o>-

-)

«:

00 0^ o o o o^

o o o o o

(Dv0*CVlO(J>NlfiC0.-v0>0tH(\jri'*<i-ir)'Oo>i^(r>(J>

119

P«6CESSCJR ACTIVITY 9N CPTJ 1

TYPE aP INTERRUPT TALLY INT/SEC

ESI INTERPR6CFSS6R ?646 6.8ISI INPUT MONITOR 1162 2.2I SI 6UTPUT M6NIT6R 458 .8ISI FUNCTieN Mf^NIT6R 0 .0ISI EXTERNAL 19126 35.4ISI INPUT M6NIT6R ( S/G ) 1472 2.7ISI 6UTPUT M€)NIT6« (S/G) 1812 3.4ESI INPUT MONITOR 1350 2.5ESI 6UTPUT MONITOR 4053 7.5ESI EXTERNAL 2468 4.6DAY CLGCK 0 ,0REAL TIME CL6CK 18848 34.9INTEKPReCESS6R 192 .4ILLEGAL INSTRUCTI-^N 2 .0GUARD MODE 0 .0FLOATING POINT OVERFLOTV 0 .0FLOATING POINT UNDERFLOW 0 .0DIVIDE FAULT 0 .0EXEC TEST AND SETS 10095 18.7USER TEST AND SETS 0 .0

PROCESSOR WENT IDLELOOPED IN IDLE ACTIVITYIDLE TIMEIDLE TIMEIDLE PERCENTAGEIDLE PERCENTAGE

45211 TIMES2555395 TIMES

297.638 SECONDS BY SIP COLLECTION291.599 SECONDS BY DRP CALCULATION

55% BY SIP COLLECTION54% BY DRP CALCULATION

SYS IDLE LOOP CT 2284437 CPU IDLE LOOP CT 270908ABNORMAL ENTRIES TO SYS IDLE 7241RT OUTPUT ACTIVITY TIME - .000 SECONDS

120

n a * # # # # tf (0

1

1

o o o o (\J o o o o o

w 1 # # * * CVJ

• o 00 o o o o o CM o o o1

1 1^ *1 oo o o o o o N o o o o o1 cy <t

o 1 x> !^ tli )^ o• P) o o O o o (M O o o o o o1 © in cc <t o o

CO 01

w • IT. o o O o o ro in * 00 o o o o s• (VI «^ U) <t

H • ?0} 1 to

1 {Hi

•< 1

1

•< 1

•

•<

01 (0

o I O" 0> tn o o O o o « o o o o p1 •* n n (D o •H

p » 1 vD * •<t * <t 00

p. > 1 (\J IH >u •< a <

a

» a ^ n to o o O o o o o o CO o »H o a <\J in >c CM u> < a ri o> tH <

a w r> ro

a

U<

HD0.H • <o in o o o o o >o o o o o C\l oD w a in oo 00 o 0> m

p a t-i in ^ oo ro p• a in 0* ro ^^ tH »

a in tvj 00 in vO «D a w <o in n

a « »H tH

z a

a

a o o o o o in vH ro o o o o oa

a

ro *

I 00 o o o o o o «^ r- oo o o l\J oo a r< o o o o o CO <c o o CD o ow a o o in o o o o wm a ww a in o OO m <o

a ou D C\J

M a {Vl *H ma

IH 1 HH 1 H

(M •* o » n K n n CM <t o PS EC OCD u u u CD t-» p CJ ID oh- * * e * u CM

OO OO .J l-l >-) hJ 00 oo Ou IJ z zn M M < M M M M n M M < H l- XC » N N N N Q p N (J •< u

a o tH D * in <C 00 0* o tH C\J D It in

a

Ha >J •J lj IJ iJ hI >-l ij i-l >-l P

P a w M u lu 111 M u U w M U w M w U4 u 0.

B. a z Z z z z Z z z z Z z z Z z z z HZ a z Z z z z Z z z z Z z z Z z z z BM 1 < <! < < < < •< •< < •< < < •< < •< < ©

1 B OB pa K n K K tc tn n n> u O o o u U u o u O o o u u u o

a >f If )f i«2

a

a

r>) o o o o o o ^ o (VI o o o o o

a # * )^ *i in o 00 o o o o o CVI 10 o o O o o'

KU

a * * # * * If h.a < 00 r- o o o o o h ^ CVI o o O o oa r-* CM (VI

IC

Pa # «^ * * * # * wa t^ to o o o o o CVI o o o o o o Ka «j o in 00 o o M

Po lu

w zIH

,-1

VIUl Ul

toa M Ha o « « o o o o <0 in in o o o o Ha n in CVI (VI >o (-1

a Ul za p: Ma H OS

• 0.

• !>-

1 hJ ID

• ZC

Pa 0^ tH o o o o o ^ 10 CD o o o U4 UJ

a <o in in tH o 0> OS

a <t <f « tH <t m 00 (0 < Ma ro CM (VI pa 03 aa It l-l Ul

W Kl-l zPJ U<z <

a 10 o in tH o o o o o tH oo in o o Ul o z PB IH

a 0> >c o ro ro in < oa 00 10 CVJ CVJ tc

a u OS Ua U' p

o: H pM z ^0. M u

CK zP 0. t-l

Pa 00 (\J >o o o o o o ro <* oo o o o o E Ui Ha CVJ ^ [0 in C o •* ro CM H m ID

a tH tH CVI Ov in lO ro PQ Ot Za OD tH iH (VI oo If) «H in tH z 0]

a Q' * oo tH tH * l>< O tH tH in U PC Pa m oa H M Pg s IS Z

Ot •<

lu

p l-l ha. IH

ZP

a t\l 10 o o o o o (VI o CM o o o o oa < p I-)

a zo:

a CD N * 0 o o o o in •* in o o 10 o p OS H1 <t ro (^ in o o o o in o in o o o D Za CM (D CVJ o o o o o <t o o o o z U. t>a ua (7- CD in 00 o tH 0]

n tH (u 111

a M z pa

a IH

1 1.) H<

lU

>H H OS

H wU lu

CVI o m to IS CO n CM o OH pq o ij € < (uoo tH p {J u o V CO « o u o ID a. pr^ It in t- l-l f H t^ <* m CVJ in l-l PD

CD oo Cu l-l l-l l-l l-l tH 00 oo ou •J z <;> z m w(C M IH < IH M M to tH •< M H t«! H z Hb. P p N N N ^4 lb p P t-i M U < u z z

•< M10 «

a o 0.

a o tH <VI ro <t in 10 00 o tH CVI 10 in w Ul lu

a pp to PC

a H H Ha 1-1 l-l <-l l-J l-l l-l hJ iJ l-l l-l l-l .J i4 l-l

a lu lu W lU HI lU lU HI HI lu Ul lU w w W wa z z z z z z z z z z z Z z z z zi z z z z z z z z z z z z z z z z M< •< < < •< < < •< < < < < •< < -< < (H

a IE IT to cn s IE PC pa to PC DC pa to PC BP

a u u u o u u u V u o u o w o o u Z

121

W JSJ lO

-< o «

tf) OD 0>

OOOOOOOOOOOOOOOO

OOOOOOOO

o n o o o o <»

o 0^ o o o o n

o « o o o o «

4-> !->

C t/1

tX) I

I/) I

: +-> ^ T3

3 o

r +-> e j::

U QJ +->

O) +->

I X >> o

a. m> XI- m

> UJ

m >

mUl I-D WO MX

s: ulU

t- o(/)

> u(/) UlUJ XD OlUl

U

< oan I-

a(/I

(\j o ns o o

r)<t(\ji^oinoO(j>(\iO'Ci*riinooooooopj rn «r u) in « If,

©

N CO lO W O00 o> 00

o» in n

^ ^ O CD

m -9 tn * <\j N« in in o lO

1^ c\j

O O O O O O o

>ot\JOlDO«oo^.(^c\J'0^^«'Ooooooo*

0

o-*r)vooiooo(7ironinf^ooo*m-<Nc\iu< in -^-o (\jinain

o o o o -•

S I—

S 4-'

o- S- +->

4-> u c rj

I- VI in

o-H(\jUt-i(Mn'a'in«r-o«c\jnu-<(yn'9in>o

-c I— cr c +->

1- s- cn X<U O M- (T3 QJ

4- O S-E •!-> QJ CU3 TD CU > 3

OJ 1— O) tc <U

OJ OJ E E OJ cr-E 3 O 3 ^I— cr o c I— (D

4^ OJ

(\J(\JCM*^-».**^^-<fNr-h-N0D(D(DCOC0CDCD

z u U U u U w U u U U uUJ l/l y; III I/) u> VJ v> 1/) CO VI in m CO V) 01 UJ LU UJ UJ UJ Ul1- D a D 1-1 i-» IH a a a a a a ain a O < (/) a CC CX Q Q Q o o a Q Q Q O a o < < < < < < <>- Q 1- >• o Q o t-i/i W <t o V) (M cy <» 3- •* «t •* * * <* O O o oIS 00 HI 00 CO OO 0 « « iC « «D * D ^ •» * <t 9 t <f <t •*

in CO D 3 I/) <0 CO OO CD OO (0 OO OO CD (D OO 00 3 3 D 3

122

><

mD (O o

z Ih

M MH CO Z0) P^0)

p 0101 ro

CO ro in>< p on n

ITr)

c cn

z 0!M

s WHe >- vO

o sain10

n n»-<

(AIN

». in Pt)

< Koe< WZ (0

»Jf o olu 01 01

t-l p H M(0 PS P a>< Q01 Rl on OO «p N * •*

01 CO OO

mH01

I HZ PIC OW

K o:

cU. sr.

©H HM t-

B H01

X ©< 0.

z

wn 0101 H

01

K UC po

V> KMP >-M h401 Z

ow

0 o•< wK Z

H HZ z01 ^o wu >Bi X)a

Z M

H C

01

W HOS UP UnJ p1-1 o<. wlb w

^— ni .

@in o o o o<M (VI <y «

H m <o inv4 «-< rn

OS 01

Ci z01 M

W MW 01

u< I

K« h

* ^ o o «10 « -•

•* <f <t P(D OO CO OO 0*H4 l-l M M <O P P P H

n <t lO 1^ 01

M (O 1/1

c o rtj OJ c Ol aO Q. +-> o 1- cfO 0) Ol 1/1

+-> C u E 4J s- +-)

c- O fO +J Ol t/l

(/» X S- 3 OlO +-> 3 o Ol M- cCL'i-

S-

pr do at ea

Q. S-O OJ 1*- ro "S "xcn OJ +J 01 +-I >i

>> (/l -o OJ C c: i/i CO1— o l-> I/) Ol M OJr— +J t/1 E OJ 1/1 ofO 0) O i- OJ ^ Ol crS- -i^ Z3 OJ s_ +-> OJ=5 OJ O" 1/1 Q. o cr OJ+J Ol 01 s- OJ Ol

s- TD o s- M-c: 1 p OJ M- O

0) 11- > M- 14-OJ S- o o OJ O S- OJ3 Q. E OJ

s- 4-> OJ S- JD s.

01OJ >> c 01 E OJ

fO XI CT C XIE O E fO o O E o c E

O o u OJ 13 13

<!

OS

+-> c (/I +J -C C +J(/I >^ QJ (/) -!-> fO01 ro 01 O l/l Ol 1/1 +-> OJH 3 E OJ o C o o _1ZC3" 1— c D- 1— 1— o. 1— h-OJ O

rj

00

S- LUXct UJ Q LiJ U- C3

> OJ +-> i+- c

+J 3 t- Ol1/1 Q_ cr -COl O 01 Ol t—3 >< S- i_

I— u Q- > Ol

•I— O 1/1

3 OJ •

OJ lO .—

: ui -c:. o cI U O 4-1

1 't- 1/1

: -O 4-> =3

1 OJ •.- E: u•r- O C

; > Q. oi S-

; OJ s- +J1/1 OJ ro^ 1—

in o00 o M

n fH 01

•<

»

HZ>u

aw

01 OS

z pID 01M <01

o a 01 oin OO in

r) (D in 0.

00 cn CO zCM CM w

oe

w PJ p11

u. pw

z•< o» M01 01

01o <

CM 00 ^ o00 o

«-« >JMb.ce

Pu•<

01

zp

OS zM M» X

° O •o o <CM (Vi M zlO in K

PjHE

H H

1— 3

C U T3 O

M- (/I 4-1 S_ in QJOJ S- 01 01 OJ S-

-O 3 O U C 01cr Q-

3 01 OJ

3C3"i— 00

ID ••-

in +J t/l

>)

O ID CO r—X! O

OJ Ol S-

E .- >,4-JO C71 ra c

I— Ol 01 i-C OJ U CT

' I— x: tn s-

— QJ. o u) s- o

"D T3 O •- OJ O

>, O Q- OJ1— +-> O) S-

r— OlM- ro roO 3 S.

cr O

3 3 cnC O ro

+J +CQ E

ro 1

—

S- >, O -o(— It- (O

n u ro T33 01

: CO -a -oCO OJ

1 OJ ro OJo c:u c

- ro O OJ1 J31 I— _c; ro CJl >)

01 3 O S-S- 01 C fO

0- -a +->

1—' c: •

ro 1/1

-. _ . ro

123

^ O

Q wh) n o

Q >JiJ m cID EC

Q PC Vi M wJ W p P p UC w 10 P M © 10 i-l

paAP AS

z»

XId

H X OS

p ae

H H< m z •< m m n n U

W u < u •< Pa. z t- SB z H z!h M < M •< u u u u UH n «

t-PQ Q <

I-D s z z z z <111 »< H H U

S- M- 1— rn

I— cn c: 13 cn o

I/) mon QJCD EE -i-

CD -O

Q- O

M- S- +-> O

O C I— -c

OJ C to r— 1

^ 13 C CD •

o o oo o o

o o oo o oo o o

o o oo o oo o o

o oo o o o o o

©

©0©©©

c <_> zEX

3 -aO r—dJ (1)

oowNtoX

CO WM

. P«Vi

00 o

&< oo

V• n

03 W(L lb

Pn (u

M a.

ts in

p » wMOP» iJ z

pw

If) XW W 11

ri ^ h4^ OO

ft- \ iJ

•0 PI •<

• <D

m

V P 4in 00

10 < •

in IK •

10 © pcn 0) CO CO H wM w < in mPu a. (fl cu »< < p in in

H H < ID p (0 aX W < K

0. M X C pu y p in 0.

< -< w in P pa

« K w pa lu

H H Pu o » wCL. 1 t U) a.

p P VI

IE

w K u P R »EE o z z ftj W

< < n X

124

N

><

CO

pHPwDOcn

(0

cn

z

KKW>

<!

H<P

WPJ

HCQ

0}

nPCO

mi-i

pp w o^ N o

M t

CQ

|£)o

s

S <>-

P P oo

oCO

HCO

pM pP oOw

H •

Hz z o

Mi-l

p P•< P

o o oo o o

(0

zPwI

<

©OMKI

z

CO

zpwi

06

©<!

H<P

W

n•<

w

OJ EN 3•I- C

O)

+-> O)

_ 'acr QJ

s- o_ o +->

CDM-ro CU 1—S- S- <o0) 0) :3> ^ o-

OJ

T3s- ro O0) Es 0)

t/1 >^+-> 13 S-

ro oo E

+-> Ol-o E

l/l

+-> ra1/1 od) o '4-

rj Oa- >7CU QJs_ +->

+J ro4- s_O QJ

l/l s-s- QJ0) ro M-J3 u l/>

E OJJ3 ro

c s_s- -tJ

oM QJ+->

-l-J

CO oOJ

<+- -a QJro U

o o XCU _i OJDlI/) o

c+J 13s- s-o >> s-Q- JD OJOJ >s- ed o

rotu +->

ro1— crx)

125

VMto

NOUw

0»»OCDO^-00^-l<^00000000000•-•00000000000000000000000000000

© o n o o (vj o<t r< O

Ifl

o<tooooooooooooo»<ooooooooooooocoooooooooooooo

© O O O <M o oo o<t \D

OOOOOOOOOOOOOOMinOOO oooooooooooooooooooooo

oini/)U)ifioooif)i^eoooooooooooor)cuoooooooooocooooooa If) <M ri 01

ooooooooooo

aiCL.B<Bui3aOHOuti.BuoHii<CLaiai(LPuDuo.OHiLPua.aiPL,a,B.p>a<0<CLiBuaia>(i<a<PL<ii<ihO4(i<Puii<a<^Zizzzzzzzzzzzzzzzzzzzzzzzazzzzzazzzzzzzzzzz

K <fi »H

<Ztt.mw(nHixxi><Qsctt.5!<aM>-iOUO'-<OQSB.<<<!WOSB5-<!WUP2

Hi<

KiHZweecetfi<fiMWWWiauinz<oszi-Hmt;MOT»it<"'^<KmmcDmiE(aiiiiauzu<'c>Me>mZ'«;M<uciupUCC»©«©©©T)'<<OOPlI.MIH< H ft.

MS:WPBt-<iaU<UUUUUUUUWWW«)WWCfeW<»WWW«6(ftt»W<6u,p<w(i,upaii!P3<!<<;<<:<<:uuwuoyyouuwuuuouu

HZ

o ©a z M< w (B

K « Z

126

i/i at4-> +J1/1 aj

S- 4->

C <T3 OJ

O -O r—

3o

4- +J i.•r- OJ

O T3

.

—

E O

-iTTtn c -cs- oQJ+-> +JQJ O-E o s_

0)s_ QJ -(->

IT! x: QJQ. 1—

fCS-

UJ fOC3 +-> Q-

3 t/J s_

o >- o ^oo O- UJ

QJ OS'o OJ S- LA)

>-(/) oo

l/l

o E QJ

o +-I ^OJ s- l-J T3<- 4- QJfO QJ

T3-c <u QJ +->

OJ 0) 0) o (J S- >1I/) QJ QJ s- !- C n3 QJ S-

S- S. o •1- QJ -l-> S- QJ

o 3 S- S_ > rol/l t_) 3 QJ Ol o O 3c QJ QJ M- S- u l-J

o o s_ s_ S_ S- QJ IT3 -a UQJ QJ 1- s- QJ s_ nD

4-> +-> to 4- 14- 3 O. l/l X3ro Q) Q) a- LJ LO

>, S- S- QJ QJ O o -!-> QJ> on D. D. Ol O COJ +-> < oS_ M 11 II II II o. r-;

X3 o >> QJQ UJ LU LU LU Ll_ QJ

rO Ol cc q: LU QJ E x: l->c o 1— tn +->

QJ o 3 o 3 D- QJ TD QJ_c C C

1— cn 1— o ID

=;

127

< w» as

M •<

HO 2

>- H4 •<

to K

in

w%MiJ

!Z•J1-1

<: 01

<lb »

U MW V)

01swOS >J

U HH BO *<K

i-l U.

D H

0) U)

OS ww >m Ma H

s:. uu

S X9 MZHI UIK IE

< Ha

w Hts

H C

oooooooo oooo*r)oooocom»<moor>(\j>o>DOO(ti\ooou>N(Dh-*inooiooo

M ZOS MC .J

Z I-l

V <

M t-i

n cH Z

f— O

. 1/1 fn1 to c, t/i 4_) .

di (>i _' c i-1 -I- X CD: r— cu +->

inioini<>ininu)uiinu>inu>ininininininintnir)Uiinininin(\i«u)iniriininininioininininintnir)ir)inin

tnnrinpiinpinninforopirjririmriioionrinrfnioNh-nriririr/rjririripironririt'iririf'i

.-(Min<t.-i<\ir)« — wr)<iniC»"CUfo*t^o.M<Mi«i*in>ot^OiH(vjri<^in\ONO'-c>jn<tiniONo-«(\ioooOi-i»<«»"Ciiww<vico<vip)nr')r)r)<t*'*<j-<*<t<t<(-u)inininifiininir)vO'00'0>OiO'0'Ot^Nt^oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo

128

UJ

;i)s

Q U)

01 UJin

<LU

_l

III

> O 01liJ K

(- I- l/l

^<:£^UJ o

_| K UJ

o "-a

O O vO

— in

n <* ^ cvj

c\j <c M —CM

N If) No ^<VJ 0> 10

* IT)

m oi *o> m oOO If) w

-< <\j CM -" in fn

^ u) CM <t n« -« cvi

n •<) <D « CM *o u} (j» o

CM cr in CO in CM

<r U> O CM

o111

10in Nino •

>0 CD

CM *

VUJin

in s* CM

oUJ

1/)

o \OO 1^

o •

vO CD

CM •*

D 0)OJ

CU>

cx

CD U03 ''"^

>^ro fO

dJ E 13s_

>> =>

O +J0)

+-> s-

> E+-> ? '

ECO s-

o

"O_J n3oo OlQ. CXUJ ^ +-> Q.

CTOJ M- 01 UJs- m

>, -aOl j3 >> c!-> J3 (T3

rOOl "O '

T3 0) S_ <U +JCU +-> s- 3 -1-

(O O) <D (OM- = 3o <D CT

1 0) TDc E Ol co E S- s_ o

o 03 _]

<* o o « o CM O OI

!->

Cooon3

OOJ >iCLcn

on

(/)

o<L> <_>

co lA

St

4- OJ

nb

>> 0)+J s-

s. 0)o {/) (/I

Oi OJx:

Q. 1—eq

+-> s-

+-> _jn3 1/1 o

CU oCL) D-a crx

CU UJ<u s-x:+-> _Jo rai/i o

Q. "4-

X oUJ

E3 <u ox: LO

-o -I-'

OJ S-s: o

o t-

in <\j CM o o o in a o s o n o CO

* m N N -00 C\J CO o IT-4 in o C3

in (/) in in «) cn (fl inQQQOOOOOzcEixizcEaiixa:ootDuooocj

-^-COvOCM-^OOtOCM1-1 n \o (M in •-'

CM in

w) in in in in in w) inOQQQQQQQoooooocio^(DiOCM'^COvClM

m >o CM in

I(M in

l/l U) V) in m in m in in l/l in Wl in l/l in

o o O o a o Q a O Q o u o Qq: It a a a a: cr CE cr (X !£ a q: (X

o o o o o o o D o O o o o o os 3 s « %

* 00 (VI (D « CM •* 0) (\l OO «n CM in 10 « CM in

CM U) CM

> >01 01C3 —I

129

>> c<u o+-> o

u10

•o<D 3 >,EE

sco

•a St nun

<u >-> o

t/) OJ t/i

+-> u CO tn0) *->

00 3 to

•D O" 010) 0) Ol 3

s- crli- <uon OJ s-

V) i/>

3 4-> Ol _1E (tj J= o

00 1— o(/) a.4-> 0) Xto J3 LjJ

Ol Q_3 -t-J O r-cj- o 1— r—<U cs- c

(O r-- 14-

(U o o+J o

>> &^OJ n3 OD un

(U

UJUl 1- •r- O3 ><-

V)t-

UJ

oUJ

K>KMce

oixQ.-**h-ino-'ooo

ro w in a) cvi * (\j

lO o m o

u) (/) «> in v) w) u) uiaoooQUOOOOOUOOOCJ

X « 2 #

>o (VI <t o « niro >0 <\l If) —

CM Ifl

+->

to13 OJ01 Ol 3

C/1 o-<^ o 3 OJ

01 i-o Cc JZ0) oVfAL

ed

OJ O) oAr

SL

U-l coc he "o

ga

"O 4-1 lO0) tn

•M OJ oo 03 i- OJo x:

+J+-> d.03 0) tn 1 Q-x: OJ o 03

Ol o-C 3 Cu to

a.x+-> •!- (-> E UJ

-t-J 03 •a0) ^ OJ3 l/l oO" 03 -t-l

Ol 3 on cn tos_ 4-1 M- OJ

1 to +-> O s_OJ o 0)+-> o 3 cn 0) 4-J

03 O- CT dr- X Ol o OJ

"O LiJ S- 0) +-I

0) 03 >iE O) >> Ol s- 4-1

E 00 s- _J -o 03f- 3 03 S- o Ol O. !-

1 03 E O o Ol OJ OC <J •- E Q_ (_)

O 0) S- 0) X X s-Z XI O- E UJ OJ cC Q.

o o

C • 013 X) -i^

0) U4-1 T3 OC C I—O) 03 ja3 Q-X c

4-1 Ol O03 •»-

_j in4-1 O cO OS_J Q- Q.O X XO UJ Ol

X 4-1 _JUJ OJ o

-C Ol<- 4-1 Q.O X

to UJ4-1 T3C S. H-3 O O •

0 3 CLE s- X(Q 14- Ol UJa J2 D-

E Ex3 !- 3 UJE OJ C•.-XI >,C E E XI-3 3E C E C

•I- OJ01 OJ X >^ ^ 034-1 4-1 E cn

I I

CO Q

s ©

© <:®

_i t-n y) u in

_i u UJ o <o a 3 I/) -1 3

to u X O CD

-D >• a _j UJ UJ ZK X o !£ UJ IJ

O UJ u > o00OJ

cr a m O < £o 2 X V) <

S- C> UJ VI 3 U_c If a UJ K O u1- a u. > X UI -1

CD 2 la o (/) -11o > U) < o 3 <o m z U X K aa. a: UJ u X _l -1X

I/) D UJ UJ n DUJt- UJ UJ z u U

Z UJ a < K a a3 a o >- u UJ Z X XU 01 UJ UJ UJ

u _i in 11 UJ a< t- UJ 3 (Z ^ z

_l U) _l 3 3 3o X UJ < o UI U zo o u Ka z a o a UJ UJ X XX m UJ 1- X X I < <

r < a 1^ z

in(/I in o

• • • •

q: •* m <> nj

UI

Q. fVJ N

n >- * UI

I » _l P) (Ji 01 OX -1 0> lO in

< in

H«

n (M

in I/)

UJ UJ

a: q:in W) o oUJ UJ K (-

> > (/) in< < UJ UJin (/) a q:

a. IX CL aUJ UJ UJ UJ

K t- Kin m m m»-t •H

13 13

UJ UJ UJ UJ

X q: a a

a u: q:

o o o oz -) z -)

»-» < <z z y

O) 01> >OJ 01

130

owK O o o o o o oo m o o o o o

h OOOOOOOOOOO<tooooooooooo

0^ (\l <u

a

hU•<o<too^oo^o<-<ooooin»«ooinoOr"00t->0000000000n00t-0b. •

V)

p•<

\ oO 00

H •

ooron(Oor^a)0'Ooroo(\jr)roir^oooo^f^»onoor)o^flD(MN(\i*oo»o-^»^o*Hh-oo(\jon*HO»-to(vioP)r)f^U)ot^moor)or)<joD<vj<jinooo>tHtvi<ti^oo<f<t(\jr)(\/or^\D4inmo<tvooo'0.-i^oo^o

0©

O o w o o o o»<orioooo»«»HOONOo oninooooo<tin(yoooo owooo<too o .H ^ 1^ (0 o -I

^ (0

0

0)

l-cn

OlUO(000«00«Oh-OOOrtCM>000>0000*0000000>0(\l<tOOOOO(\JOOO<tOOO<t-(\I<\J\OOU)00

©=£0*tn(?rorno><joio<vi»<<trooN»<>oo)r)io<Mio«i/iifir)vor^^onina>-<>OiMino»<<t-«^rooo.*»<oocotoin<0'0

o z 0. 0. 0) 0- hI to H VI 2 IS VI w la! 11 P H z a Qo M o o u M H Ul M » K K PS K VJ U u Q < VI Z ^- HI O e- H u q: hI pu. >• B. X n a R » w w H OS 2 < 0. U B. D u l-l l-l w M 0S M a (J z 2 M lU < U] 2 u u o hI pj €> HI

< HI 01 lu B z •< <: B. y 2 u M i-i p « 2 2 z lb u. >J lu VI VI u z l-l UJ u u o o s 2 < 2 Z X z <; HI M K lb 2 MX Oi M X s s » U) UJ < w S >- >> >- lu c •«; « < a, c < < < u «: M M M m lb VI Ul VI Ul < VI V) HI Ul lu VI

w p 01 Ul (0 Ul (A (0 VI 0) IS P u D m W V) to OS M b< Ik u. lb p M o o VI < l-l z Ih lb lb lb (b U u u U P U U lb u W u

wlnNc^^o|^^a)r)l^<\la>•H<l<vl^cM^-l^>^o^ coiciniEwoDflon^ nriio <fvc« 01 «o<riOtit^

Hz 0

r-CT'<\itnt^rot--<DMi^(viHi<tCT't^w<\i in«-i oi»"»< 01

-I N (\i * in oi

r) u) o >o m o oID 4 in -< w

10 in

»^<\j.-<i-ioino^o(\j»ooi<oi<DPi4"0^o(\iinrotf>r^a3^^(sj.-*vD(^«o<\i(\j'00inr>ojw»o

01 -< 4w «eyoDrtM<\j u)>0(Mi-i (MM^ioin

^f Hi<\in<ti()<ONooo>o.M*io^o)O^Oi-i<\i(oniO(DO>Oi-<cyr)iO<oo>oro<fuiit(00>Oi->woo-<cviri«inioNo>'"r'-•««-i-<,H^c\jw<\i(\in)ninn*«)-«it<i-<t<i-ininuiifiininin>oiCioiOt^NNl^r^i^Nr^N(Ooo

131

oooooooooo^oooooooooooooo&0000000000(\)OOOOOOOOOOOOOOC.....t.. ...J(E

Vi

J<H«

ww

o^ooin«oooooo'Ooo^ooino^oor)o<ton(Oi/)inooip*oooicot-iu o»<ooin

I" ^* w «. . Z

•H « •<

<D Ccr 1/1 "D<U QJ -r- OJs_ 3 CO 5

cr (U ocn s-•,- S- 01-C OJ o c

3 -Ecr CD

fo (13 -i<:

s- t/i o^ <u tu+->>+-> 4-J

r— (T3 l/l

LO o TD CU4-> I— C 1/1

CD fO O CU

E S-CDTD (1) u<L C S- C

cr (/I

q: t/)

cr re -M cr

fO -r-

+J S- -MOOJ _Q OJ S-3 3 CUcr dJ crjD

c>, l/l

+J OJ +-> <u•r- s- •!- cr> 3 > lO•I- l/l (- s_+-> (U +J (Uu QJ u > _=c e: =£ =t h-

oooooooooocooooooooooooo

ooooooooooNOOooooooooooo(\iin

4 (M <t * -1 *

«; I z c w01 «: < c u aU CL B. U X U>

t < < M < H« O Q S Q h,

e 0. tn M MZ Z 2 2 S3:> ^ c u

i O U O

n ijBh w >-

•< 2 U.

S ^ »U] V) Ul WMJ^iMt-HMtHt-WUlOW

Vwin

0- iC CM lO r; n (Vi o O CJ in (\j n CJ lO o <* (Vi

O) (VI (0 iC in 10 * oo <o

OJ w CJ <M * t~- min oo ff

olu

(0

\Ol n 00 CVI vO <Vi o n * N N <\j in w m <D o lO (VI *H

01 ro o •O fH CVl (VJ

ri w < {VI oo oin m (VI

to

J•<

t- m Pi <t (0 o 1/) 10 ^ <o n Ul 00 o CVI HCO (D 0' 01 0- o o o o o o CO CM n M m m n n n It * C

H

-^-p

—

CU -C ^+-> 3

O -D

CJ) <U l/l +J

U ro •'-

- -r- OJ 4-CD "D -C 1/

S- C S- <ur- QJ >

_ > C•I- Qj O

S-_ cu ct djcI X: Q JD

sec •

M O O •>- -Ml/l 1— «- S- COJ +J +-> OJ3 O U l/l EI

CD 3 S- OJS- l/l 4- O l/l

OJ OJ OJ

(1) '.-

^ +->

5 U

O OJ </) o o> CD O +J

u -o -o+-) -r-

OJ (J

132

3 O)U •!-

o -utC -1-

>(U -r-

S- +->

+-> o_rO >-<

+-> Ul

QJ E OS- fT3

O S- t-O

Oo s- +->

+-> Q.o

O) Ol +->

s- cn c03 n3Q. S- i_

E OJ a.O >O cC

iUl

•

O >lJ 03 lf>

f\J OJ

<( CT CO D• • • (\)

in Ol

J

<

H<!

t5

M

m

tn (fl

QJ 1

+J

COEO)+->

t/1

OO

(/> c:

-l-> +->

ra3 sin QJOl S-

QJS

!-> Ul

QJ ZD-o D-O

©(M O « vC CO <y <c

U) « (J> ^

M Ul

w

» u

<

<\J O lO CD (VI vO

in w h 0^ w

p c MO U o Ho z u O

2. K H ->! w CM * o o >0

M Mj Ul 0. m Z CO CO f-( <o « f*Dfi y CO 05 c> HI * <t <* * P

Ifl T < o in in in » p: w rH CO 00 a; CO

P- u fp H u • • (- H EC M M M <Ul W •< «! 2 CO CO U) U) U. P C R

u< z z M OJ CO OJ w >- ^(h HJ M M H K H w UlID <! CP (D CO

w UJ Ul Ul UJ © P p ro

?5

C :UN UNzDzpzp

U; U7 Ul Ul

M C>« Oi W K HH

I-I

<i U MID U

oH U » >-) M

h) •< oO bi < UjM O w

p «:

3 ID UJ

HI Ul

H H HUl W U>- HUl W Ul CK

1-1 a.

cm Ul

i-i

ID <; I-)

H V y •<

133

THIS SECTICN GIVES THE NUMBER 6F ERS BY TYPE FflE (3NI,Y THOSE REQUESTED

BASIC EXEC PR'S:

This report may be usefulin explaining why EXECoverhead is high.

# ER TALLY ER RATE/SEC. ER GR0UI

1

.

I0S 3128 5.8 22. I0IS 89 . 2 23 . lews 21 540 3? . 9 26. WAITS 1907 3 . 5 37. WANY? 937 1 . 38. CAMS 66 . 1 1

9. EXITS 2052 3 . 8 611 . Ff^RKS 253 . 5 6

13. REAPS 1 072 2 . 0 4lA. PRINTS 2531 4.7 415. CSFS 561 1 .0 716. leAxif 128 . 2 218. 57 . 1 1

19. TIMES 165 , 3 1

21 . lexiJt 38 .1 2

23. IIS 4 . 0 6

26. FITEM? 23 . 0 7

33, MOTS 1 . 0 1

35 , MCf RES 13 .0 1

38, CMTS 2 . 0 539, CMIS 2 . 0 5

40. CMflS 1 S80 2.9 543 . CMSAS 2723 5.0 544 . TDATFS 1 44 .3 1

, TWAITS 34856 64 , 6 351 . 16 . 0 1

52 . PCTf 27 . 1 1

53 , SETCS 4 . 0 1

56. APRNTS 51 . 1 465. TALLP 3-»

. 1 666 . TREADS 1 .0 4

68. PFI « Q . 0 7

69 . PFS« 4 . 0-

72 . PFWL« 9 . 0 773. LaADS 1 47 , 3 1

76 . FACTLS 1 1 1 . 2 7

85 . MSCflNS 1 . 0 787. SWTCHS 47061 8''. 2 692 . AWAITf 5 . 0 393. TSWAPS 1 . 0 795 . PRTTNS 3 . 0 496 . ACSFS 1 5 . 0 799 . FACTTft 1 49 .3 7101. PNCHAS 1 4f, .3 4

1 02 . NAMES 6 . 0 61 03 . ACTS 3248 15.3 61 04 . DACTS 7567 14.1 6

1 06. CRE L$ 2 . 0 5111. PSRS .0 1

112. BANK3 9 .0 1

113. ADEDS 12 .0 6118. ARE ADS 2 . 0 4

1 20. ATREAD 45 .1 4

121 . LINK? 8 .0 6123. EXLNKS 4 .0 6

1 24. UNLNPS . 0 6125. RLI STS 8 . 0 6

1 37624 2 54.8 8

ER GPflUP

BASIC EXEC ER'S:ALL EXEC ER'Sle RELATED ER'SER'S THAT CAUSE ACTIVITIES T0 WAITER'S ENVeLVED WITH SYMBIflNT ACTIVITYCOMMUNICATIfN RELATED ER'SER'S THAT START i. STOP ACTIVITIESFACILITY INVENTflHY RELATED ER'S

TALLY ER RATE/SEC. FR GRflUP

13762^ 254.9 1

24923 46.2 2

37705 69. R 3

3351 7,1 4

4309 8.0 5

65285 120.9 6

897 1.7 7

134

pJ «C In

(- o. I- Z ^t~ < Ui

c cO X

o © o

o c o

o © o

IC oas © in

o OS oin U MK B. Ul

KJ K 10

< D Wu. s<c 0

H 2: wOS

UJ wO< M iuSi l- 2w M

•< H<

0) Ecn s-fO CD

3 «4_> _Q OJ

000

000

©

it

0

o <u

0 0 0 0 0 0 0 •* 0 • N CO

m O'

1 a

0 0 0 0 0 n a 0 CD 1 <}•

rj 0'0-

0 00 0 CD 0 a 0 1 t- CO

<t rn CD cy r> I n N1 0'

0 OD 0 in 0 o> CO 1 d

m ^ (VI r> 1 o> n1

0 IT) m « 1 O'

1 •

CO 0 ^ 1

CM n ri IX) n 1 m ID

0 <D CD <» >D CO o> 0 lO 1 <t <t

(C 00 0' CO N t- 1 0>

•0 0^ * CM CM 1 •* <*

0 00 0 0 0 10 0 1 CO

n.1

in

1 1

1 Q1 Z >

•0 1 < Mu u u a s l-

\ N 1 HI •<

1

(M

1

n

1

*

1

in

1

•0

1 1 1

1 Q1 U

l-l

ss

Q Q a 0 0 0 • < sS S X s s s S 1 i- ww w M M u w H 1 C 0

H Q 0 0 0 Q 0 Q Q P 1 h <

135

»b

PW

wDau

w

oz

<;

I"©H

MJ

O<!

h

w

«

wp

o *O It «

(VI «

o oO O• H • *

O

rM

ID oW !D OW H •

OKUH MP£ Oi W

w H 1

w (0 1 o If)

< n W 1

P 1

O O 1

H z M 1

W 1

M c o oU a o o•< M o» H • •

M M> <! H<

in 10

o o 1 o o• • 1 • •

1 oR o

o o 1 o o1 • •

1 oe o

o 1 o• t 1 1 •

?f St o(VI 1 (VI o

»-(

o 8• • 1 • •

in 1 in in

1

o a

c •1 •

1

1

I in

I QL)

I .H

s

I

I

I

I oI oI y>

I •

I

I

I

z; 1

t PB >P 1 u M< I H

1 < •<P a •J

• w R

u a <! PM H g H Ua << a ou m a •<

pmpwu

•< COH HZ z-< pP

u

<MHMz p

zR

z >w

z P

CV/ O CD (^ O 0^ (7»

I/) ro ,-1 in

(vj ro <t in »o r

136

min

pZ KC W 0t) >

Wa •< Na. p M

0> m1- f- wU K w m u.

P •< ss

H w •<

w w < n

a Nw (- K >H •<

t<: »>: t<; t<:

o G o < n

o o o <» *

M IK M o: us

o o o <\i n

o o o n nm o

M V K « Z» U M C W

» U K

U U PS IK lUM M W W 0!

S P U. b< B<

0 o w wU U Qi OS Cas IK CL. O' z1 I I I I

o tH w r) <»

«

u. a:

N MMCO 0'

n > •*

u l-

>TE

•< OS

T>

Z Mu OS

a CL

o•< CB

»

</! c•< cn

»TA B

111 oK pO (VJ a

U H z<

© % ©Z

M (n

OS <0. ?

PS MPS

Hc IE

J H

M P4CO Wl

D p< <0 uw lUn n

p W uu w

t hJ

u M< <b. vO ^

M <* tvj

w < mla! >in < Ul

< <H

>w < mUl Ul

wai mH

NM<n 4

D

s © ©

©

M Ob. OS

u.< p.

a• U)

Ul<H

ow W

x >< <

Ul i-

e HZ (1. 0)

P IX

<! Ul ••

c a < Ul

t w h <Z Ul »

« o ~Ul W H

h < K ^M H W U.

U. PP OS >J

C W C iJ

e- a •*

IL) b.

o P o >•

? b; iij

M r » tr

eK u. uH C B (1,

wUl H^ M

Ul < u.< s.

H Ul H

w PC

tt X H

b<

NMUl

(«:

MZ<;

IT

IP b.Ul >

•<

z Ul t Ul

< a Hp % t Ul

w a M MN M 1- K

wUl <a

g(t < oo t> H Ul

h N wM a

S p, Ul o MC P <\l

u w>

U 0IE HH W

•I-

5 4- c

> CDOJ 03 fT3

o .— co <u

CO U -Q

D- X "D

e: -c

V > 01

Z w z •< Mn > b.

OS 1- •< 1 Uw z ap z cns zo B

f- (-

aV H <PS M IU

0. Ul lb

at

Ul M f o Ul <lb evi u

a IB uiU <; Ul H

o z p «>

t> >0

o < < ri < pz H H o <t b. blH Ul t> O h

(D J p •«

© >J < w <) >-i

i-l P< «) Qi za H «; p

lb Ul a oUl o

u H •J <IS P. <

X pw <H <t H w

0' Ul n< r" IK

U mH •< >M CJ

(b < oP w

I-l B< o HH Ul

h<

H U<

Li- o S- >

137

wQ<

MKm

wo<

•J

uuDw

o

wQ"<

.J

wwpi

uu

w

rn r) m

wh

<

-<

o

uM

P< In

»Ww

•4

^ CM «M < u»

WhJ

CI

138

0

© sin

£ ©

oo

(-

»m in

OS

H o oai w

LEA

0z

=t <_) m 1— 1— 1— 1—

c Q1

Q Q ea; «^ o o1 1 1 1 1 1

ss1 1 1 1 1 1 1

•< •<

K P o0 H P u Hw !s a CO

» M lU s \ \2 0. o o 0M CO n w H0. D o x 0 P S5 C

H (14tn CO iZ wP o H w :s CO

CO w »<: A. t<: 0. 0.Wp CO CO m CO CO CO CO

w !Z < P -c D DPPK W H CO H CO m Si, XU 0. \ \ \ M CU

in CO CO CO Ui M a. 0.

V u u CO tn CO

CO w m w < B »< CO cn CO H H cn CO

n Mf- CD m (\i o (? (\i in

o in o O ft oUl >< • • •

•< 0 N r- in

m ;e 00•<

H Min

in <!

•* H

©

PU W ou

< n •<

sin p (0

zm wta PC

H a i--

D ^J

H K iJ

<!

11.

CO SP KS; 15

HlA

CO M cu

p cn 0.

CO <!

H ©H

b.

55 c Pw

w OS pH w wP m wD z

p0. z CO•<

» cIfl z N

MO>0 at 0>

Ifl oa

© H ©nIn mtn >" KM <

2

a. M» <m UtB CO IE

H 1-

MZ

IK

0<

^ z y* IE M WCO H s cn

N \ WH, M W Z U P P< tn a, n ID m m? © © s cn a. B.

cn o o \ N B, 0,\ cn ifl < <cn Ik u. a. Oh ^ »- ©©•<< cn cn

© » &M t>i M CO cn M M

ro 0^ 0^ 00 Ifl w rj in CD o

l^ocvi.tinuiNiT'C^or)*^<looo>o>Cf'C^o

NrtycvjoaCMCM

<-OD(M*i)OCO^ritHO^•HMCvjcvm^inoi

C' ©©©©©©©©©HHHHHt-Hl-l-'l-'o-*<o(vi\r>oooiont-i

rH ri (\j (M m n- in

o

wtc

H M,J

iH ID

m M1/1

z CO

© ©a.

p CO

o MMX s-111 M

t"© Pii<

Hw pP ©

< Wp a< p

a. U4

P < Oi

z a<

in

C5 cn< zM a'

o a)

z «;

aj cn

CO

?; <© p a)

a. a]

< mp Hw> z a.© < ©K

m Mac tu zZ ©

CO © zMH a)

EC cn «U < wM u IE

s »j ^in

a^ < zz ©© CO H

ajCO ? HH ij P© z

CO z u< M

z>- M

p hJ <!

up < Oi

z P Haj cn Wa. pcn a.

P cn MCO P

EE Z«. P

139

wJw

oU < 04 o O uw € M M M >J U u fa fa

CO to to to M to to

w >< < •< < -< < < >< < < M Hft H O o fi p p P p p p P P

p

KH w»5(ti

h o fa Ul! fa fa (0 u; Ho fa < to fa w z M

a ft; < o < fa I-l <10 OS u s 4! Pi M pa PS fa

u o fa <! o fa fa fa Hz a, to fa fa ID fa D

to a. z M u o ft; ft! IE o \ U< M c fa < fa fa fa •< fa ft! MM • fa Ui ft! p: •0 ft! <; •< faW • o to fa H fa o fa < fa «0

"< ft; (/i 06 < 10 < P to

H w ft! ;e Z ft! M to

H ft; fa ft! fa 10 £> to "< fa to P <a< (1-! ft! Z z HM M Qi o < fa fa s t •< fa

M fa < 10 10 ID 10 fa

u u w Q P ft! ft; ft! ft! M fa HCO w o < a < •< fa fa fa fa c fa

[Ll fa fa IT 10 o© o z fa ftS Hi CO to to to 10H M TV fi to P p u to z PK <! © ft! < < z z fc fa z<! fa z z 10 •< < c ID M> « M H fa "< < fa n P H fa

(0 uw

H in I/) o ro t-< n o ct in r- 00 <0

H KUi rH ft (VI (VI

(fa Dh•*

nPfl

Hzw

to in ro f-fl ro in ^0 in 00 (0 *t oCO PJ »o <t vO ro ro o roM (VI 0^ CM c^ (^

H oH ro (VJ

>< ro (VI

CO ITH

%

§M

)-) <•-) (t)

ID(fa < o c« w p z to (0 10

Q < < to M Mft! 10 P Z Z V C fa

tc w Oi •< M hJ Hi •< fa PQ UJ w !X!

H ><; u V) fa M z fa fa ID z <; < M fa

u fa M M ft: PS i-l ft! p P P Z

cn

fOI C7)

o <u fO >S- QQ 4_>

1)I-l CO i_(/} QJ Q jy

c: q_ rv</> o 1 1 1

o QJ COCD to ,—~ QJ

d) *o Q r*i

4_> • r~ i—

I

•r- +-> >fO Q 4-* OO CO *r—

> 1 c > QJdJ O Q O 't—

NX o OQ) o CO O(/I ,— c QJ fO

CD rn s_i_ i_o "O S- Q I/) QJ c~<+- "O O <C

M— O QJCD 1

^ CU <C QJ O •MCI 1 1

1

1

—

"O Q SZ -M1 o fO o "O

to [-H CQ E CO (/)1

1

fO QJo C 1

—

Q. (Jo n d) CO s_(/) ec CU QJd) -Q Q CO CO

CO1

—

CO • QJ CUfO CO r- £ QJ

o (— £ 1— 'r- r-

X j_ oCL) t)_ c i-

Q_ O CD •

o X to •1— Q QJCD S- c/) S- QJ -r— QJ

o o •p- Q_o fd to CO fO

E ri q; QJ M— QJt_ o "O o (/) O _QQ4-> o CD QJfO cc QJ C(J Q fO 4_> .r- fO

Q CQ M- "O "O<D (J fO cc~ QJ QJ •

fD QJ -QO^

CO r*> (/) C t—

'

fO O QJ •1— Xo >^ CO

i. <t5 QJ *r- QJ CDQJ Q SZ

c i_ (/I to "O o>^ CD "O CU t/) QJ QQ O j_ "O 1— -P CO £Z

o tl— r— 3 Q QJ fOCU Q- Q QJ •r- ro O

4_) [ [ ) "O "-M TQ +-> CT) CI X

•r— I-LJ

M— M- Q Q -r rvO O .o Q r— Q- Q_ ^

CD O O QJE to Q_

O OCO QJ ^v

cn (C CD 3 o 1— -fc^

O J3 O +-> O O r— "O 1— 1—

1

S- n3 o 03 CU O QJ O 3Cl O) a ro O

f~ Q-) ^ .j > +-) 1—

H

1— 4-> c~ cooQ Q d^ fO fO o oS- O CJ I

O E E C r— o oO >, 1—

1

O S- ra QJ Oo s- o 1 > •r— o s- l-H

o M- £2. •r— +-> O) 4- CD +->

O) E S- res +-> •r— t- s- o QJ C QJ "tO-

SZ CU <u U O CU _c: o _c: o1— E CU 00 1—

1

O D- Q. 1— CJ h-

cii

a; CJo 1—

1

C_> Q-_i t/O 1—

1

CJ<c eC < cC 1—

r

oQ Q Q O 1—

1

TEST AND SET SUMMARY

EXEC T/S T6TAL «

SLEXl " 18433TSSWL • 150,X • 32IE - 198TSASA - 1

JK - 1

MB " 13PCFTS - <508

KDACT - 1

SLKDNT • 23TSSLC « 33TSACS - 2

TSCCIL - 5

TSCMAP - 21TSNAMl • 12ESITEH " 292

TOTAL TRACED T/S « 19^90

21199 USER T/S T6TAL - 0

^For both processors~

Common Test and Set Cells

SLEXl: Protects the switch list queues.TSSWL: Protects information in the switch list entry.SLRUNT: Protects total accumulated run time in the PCT,

IE: Protects ER to CPOOL$ and CJOIN$.PCFTS: Protects granule item.

ESITER: Protects ESI Activity Termination Queue.TSCMAP: Protects demand/batch sharing information.

141

"OOJ-M

.c ^-•<- oE

Z P iz; Sz; ^ 2:

W W W w W W >,tswwwww xn rooj

\0 o <t a I o ^—<t vD i

<f *-j 00 O I N• (M • • • I •

. I o<J- I

W I (M

IC O © P MU. >< M

O p p pq>< O < WM (Li P PU K <D3 n < K M H09 CtS M Qi ^ O(u IX IS (u ID H> > >t O W IE. Z pZ > <! 5

Z < ID !Z <!-< -< P «P ^ W C

P W <; p Hw H w <: wH < P W tc««; w w <j Pi HM w h (t) u« O •< P&

u » u wCO 0:; %M W U U H

O tsl 05 fH

<! w <! pM W ID l» C>

U U 0!

< 2 H <a, C! <! 0. H

a- H Da. 2 M QhM ft W U HM c s <! a

C5 « ID

W U << Hn i-i PK pu

H <1 IL TD H

P wID M Pe W (14

M tc ID pn BPi H O H HMa, p p p p^ c © c

(H H H H MCO W W W PC

U (1:1 (I! (i^ (4H a< pu PL.

(/] vj c/i tfj v;

M M M H M(C n PQ [B P3

H H t-< e

c» o o o o2: z z z zM M H M MOQ W K p; pe:

D P P P DP P p p p

142

NBS-114A (Rtv. 1 )-77>

U.S. DEPT. OF COMM.BIBLIOGRAPHIC DATA

SHEET

1. PUBLICATION OR REPORT NO.

ms SP 500-34

2. Gov't AccessionNo.

3. Recipient's Accession No.

4. TITLE AND SUBTITLE

COMPUTER SCIENCE & TECHNOLOGY:UNIVAC 1108 EXEC LEVEL 32R2 Performance Handbook

5. Publication Date

June 19786. Performing Organization Code

7. AUTHOR(S)

Capt. John C. Kelly and Lt. Gary P. Route8. Performing Organ. Report No.

9. PERFORMING ORGANIZATION NAME AND ADDRESS

NATIONAL BUREAU OF STANDARDSDEPARTMENT OF COMMERCEWASHINGTON, D.C. 20234

10. Project/Task/Work Unit No.

n. Contract/Grant No.

AY-505-080-Armv12. Sponsoring Organization Name and Complete Address (Street, City, State, ZIP)

U.S. Army Military Personnel CenterAlexandria, VA 22332

13. Type of Report & PeriodCovered

Final14. Sponsoring Agency Code

15. SUPPLEMENTARY NOTES

Study performed and report prepared by the Federal Computer Performance Evaluationand Simulation Center, Washington, DC 20330.

16. ABSTRACT (A 200-word or less factual summary of most significant information. If document includes a significant

bibliography or literature survey, mention it here.)

This is a report prepared for the Army Military Personnel Center (MILPERCEN) . It

describes a set of hypotheses for evaluating the performance of UNIVAC 1100 ComputerSystems. The hypotheses specifically apply to EXEC Level 32R2 operating on UNIVAC1108 Computer Systems. Attempts to apply the guidelines to different UNIVAC 1100models or to different levels of the EXEC may lead to erroneous results.

The report contains sections on each major complex within the EXEC. Each section(1) states the performance hypotheses associated with that complex, (2) explainshow the available measurement tools may be used to determine if the hypotheses aretrue, and (3) identifies which performance parameters need to be changed to correctthe problem situation. All of the performance hypotheses discussed in the body of

the report are collected together in Appendix A. Appendix B summarizes the

performance parameters and Appendix C contains an introduction to the SoftwareInstrumentation Package (SIP)

.

17. KEY WORDS (six to twelve entries; alphabetical order; capitalize only the first letter of the first key word unless a proper

name; separated by semicolons)

Computer performance management; measurement tools; operating system performance;performance guidelines; performance measurement; tuning guides; UNIVAC 1108.

18. AVAILABILITY Unlimited 19. SECURITY CLASS(THIS REPORT)

21. NO. OF PAGES

1 !For Official Distribution. Do Not Release to NTIS

UNCL ASSIFIEDT44

1X

1Order From Sup. of Doc, U.S. Government PrintingWashington. D.C. 20402. SD Stock No. SN00irP03

Office 20. SECURITY CLASS(THIS PAGE)

22. Price

1 1Order From National Technical Information ServiceSpringfield, Virginia 22151

(NTIS)UNCLASSIFIED

$3.25

USCOMM-DC 66033-P7e

ANNOUNCEMENT OF NEW PUBLICATIONS ONCOMPUTER SCIENCE & TECHNOLOGY

Superintendent of Documents,

Government Printing Office,

Washington, D. C. 20402

Dear Sir:

Please add my name to the announcement Hst of new publications to be issued in

the series: National Bureau of Standards Special Publication 500-.

Name

Company

Address

City Stale Zip Code

(Notification key N-503)

•U.S. GOVERNMENT PRINTING OFFICE : 1978 0-261-238/143

NBS TECHNICAL PUBLICATIONS

PERIODICALS

JOURNAL OF RESEARCH—The Journal of Research

of the National Bureau of Standards reports NBS research

and development in those disciplines of the physical and

engineering sciences in which the Bureau is active. These

include physics, chemistry, engineering, mathematics, and

computer sciences. Papers cover a broad range of subjects,

with major emphasis on measurement methodology, and

the' basic technology underlying standardization. Also in-

cluded from time to time are survey articles on topics closely

related to the Bureau's technical and scientific programs. Asa special service to subscribers each issue contains complete

citations to all recent NBS publications in NBS and non-

NBS media. Issued six times a year. Annual subscription:

domestic $17.00; foreign $21.25. Single copy, $3.00 domestic;

$3.75 foreign.

Note: The Journal was formerly published in two sections:

Section A "Physics and Chemistry" and Section B "Mathe-matical Sciences."

DIMENSIONS/NBS;This monthly magazine is published to inform scientists,

engineers, businessmen, industry, teachers, students, andconsumers of the latest advances in science and technology,

I

with primary emphasis on the work at NBS. The magazine

j

highlights and reviews such issues as energy research, fire

I protection, building technology, metric conversion, pollution

I

abatement, health and safety, and consumer product per-

I

formance. In addition, it reports the results of Bureau pro-

• grams in measurement standards and techniques, properties

I

of matter and materials, engineering standards and services,

Iinstrumentation, and automatic data processing.

I

Annual subscription: Domestic, $12.50; Foreign $15.65.

INONPERIODICALS

1

Monographs—Major contributions to the technical liter-

ature on various subjects related to the Bureau's scientific

I

and technical activities.

'j Handbooks—Recommended codes of engineering and indus-

trial practice (including safety codes) developed in coopera-

jtion with interested industries, professional organizations,

!and regulatory bodies.

ISpecial Publications—Include proceedings of conferences

jsponsored by NBS, NBS annual reports, and other special

I publications appropriate to this grouping such as wall charts,

ji

pocket cards, and bibliographies.

[{Applied Mathematics Series—Mathematical tables, man-ij uals, and studies of special interest to physicists, engineers,

1chemists, biologists, mathematicians, computer programmers,land others engaged in scientific and technical work.

National Standard Reference Data Series—Provides quanti-

tative data on the physical and chemical properties of

materials, compiled from the world's literature and critically

ijevaluated. Developed under a world-wide program co-

jiprdinated by NBS. Program under authority of NationaliStandard Data Act (Public Law 90-396).

NOTE: At present the principal publication outlet for these

data is the Journal of Physical and Chemical ReferenceData (JPC'RD) published quarterly for NBS by the Ameri-can Chemical Society (ACS) and the American Institute of

Physics (AIP). Subscriptions, reprints, and supplementsavailable from ACS, 1155 Sixteenth St. N.W., Wash., D.C.20056.

Building Science Series—Disseminates technical information

developed at the Bureau on building materials, components,systems, and whole structures. The series presents research

results, test methods, and performance criteria related to the

structural and environmental functions and the durability

and safety characteristics of building elements and systems.

Technical Notes—Studies or reports which are complete in

themselves but restrictive in their treatment of a subject.

Analogous to monographs but not so comprehensive in

scope or definitive in treatment of the subject area. Oftenserve as a vehicle for final reports of work performed at

NBS under the sponsorship of other government agencies.

Voluntary Product Standards—Developed under procedures

published by the Department of Commerce in Part 10,

Title 15, of the Code of Federal Regulations. The purpose

of the standards is to establish nationally recognized require-

ments for products, and to provide all concerned interests

with a basis for common understanding of the characteristics

of the products. NBS administers this program as a supple-

ment to the activities of the private sector standardizing

organizations.

Consumer Information Series—Practical information, based

on NBS research and experience, covering areas of interest

to the consumer. Easily understandable language andillustrations provide useful background knowledge for shop-

ping in today's technological marketplace.

Order above NBS publications from: Superintendent ofDocuments, Government Printing Office, Washington, D.C.20402.

Order following NBS publications—NBSIR's and FIPS fromthe National Technical Information Services, Springfield,

Va. 22161.

Federal Information Processing Standards Publications

(FIPS PUB)—Publications in this series collectively consti-

tute the Federal Information Processing Standards Register.

Register serves as the official source of information in the

Federal Government regarding standards issued by NBSpursuant to the Federal Property and Administrative Serv-

ices Act of 1949 as amended. Public Law 89-306 (79 Stat.

1127), and as implemented by Executive Order 11717(38 FR 12315, dated May 11, 1973) and Part 6 of Tide 15

CFR (Code of Federal Regulations).

NBS Interagency Reports (NBSIR)—A special series of

interim or final reports on work performed by NBS for

outside sponsors (both government and non-government).

In general, initial distribution is handled by the sponsor;

public distribution is by the National Technical InformationServices (Springfield, Va. 22161) in paper copy or microfiche

form.

BIBLIOGRAPHIC SUBSCRIPTION SERVICES

|irhe foUowing current-awareness and literature-survey bibli-

liographies are issued periodically by the Bureau:jCryogenic Data Center Current Awareness Service. A litera-

ture survey issued biweekly. Annual subscription: Domes-:

tic, $25.00; Foreign, $30.00.

pquified Natural Gas. A literature survey issued quarterly.

|\nnual subscription: $20.00.

Superconducting Devices and Materials. A literature survey

issued quarterly. Annual subscription: $30.00. Send subscrip-

tion orders and remittances for the preceding bibliographic

services to National Bureau of Standards, Cryogenic Data

Center (275.02) Boulder, Colorado 80302.

U.S. DEPARTMENT OF COMMERCENational Bureau of StandardsWashington, D.C. 20234

OFFICIAL BUSINESS

POSTAGE AND FEES PAIDU.S. DEPARTMENT OF COMMERCE

COM-215

Penalty for Private Use, $300

SPECIAL FOURTH-CLASS RATEBOOK