Sheikh K. GhafoorDepartment of Computer Science

Tennessee Technological UniversityMay 05, 2010

OutlineOverview RMS for adaptive parallel application

A model for adaptive parallel system

ConclusionsFuture research

Autonomic Computing

 ”What is autonomic computing? It is the ability of systems to be more self-managing. The term autonomic comes from the autonomic nervous system, which controls many organs and muscles in the human body. Usually, we are unaware of its workings because it functions in an involuntary, reflexive manner -- for example, we don't notice when our heart beats faster or our blood vessels change size in response to temperature, posture, food intake, stressful experiences and other changes to which we're exposed. And, by the way, our autonomic nervous system is always working”

Alan Ganek, VP Autonomic Computing, IBM

Research Area

Autonomic computing -systems that adapt themselves without human intervention in response to unpredictable events and data Adaptive hardwareAdaptive OSAdaptive system softwareAdaptive applications

Demand for Computational Speed

Applications in science and engineering are computationally intensive

Continual demand of greater computational speed from computer systems than is currently feasible

Areas requiring great computational speed include numerical modeling and simulation of scientific and engineering problems

Examples Weather forecasting Manufacturing and design Modeling large DNA structures Modeling motion of astronomical bodies Problems in particle physics

Classification of Parallel Application [Feitelson and Rudolp 1996]

RigidFixed no. of processors defined by users

MoldableFixed no. processors defined by Resource

Management Systems (RMS)Evolving

Processors vary during execution, initiated by application

MalleableProcessors change during execution, initiated

by RMS

Adaptive Applications

Evolving and malleable applications

Change resources during execution Characteristics of Adaptive Applications

Capable of expanding or shrinking Capable of communicating with RMSCapable of negotiating resources with RMS

May have resource utilization constraints Consist of phases where adaptation can occur

Adaptive RMSCharacteristics

Handle additional scheduling event Release or request of resources from running

evolving applicationsScheduling may involve rescheduling running

applications Request for change in resource allocation Additional degree of freedom with malleable

applicationsNegotiate resource reallocation with running

applications

MotivationAdaptive applications promise

New classes of applications driven by unpredictable data and events

Autonomic computingImproved system utilizationBetter application performance

Current job schedulers and resource management systems are unable to handle adaptive applications efficientlyLack of infrastructure supports Absence of large number of adaptive

applications

Research Objective

To investigate the characteristics of adaptive workload and the RMS, and their impact on system and application performance.

Research Methodology

Investigate through prototype systemInvestigate by modeling and simulation

Research IssuesResource management system (RMS)Programming model for adaptive applicationsScheduling

RMS Requirement for Malleable ApplicationsNegotiate with running malleable applicationsAllocate/Claim resources to/from running

malleable applicationsScheduler

Consider running applications in addition to pending applications Choose candidates to allocate idle resources Choose resource preemption candidates

Resource Negotiation Protocol

Different communication scenarios are possibleSupports all possible negotiation scenariosSupports multi-round and multiple resources

negotiationsThe resource negotiation model consists of two

partiesAny party may initiate the negotiationNegotiation is done by exchanging

Negotiation Template(NT)

Scheduler

Negotiation Manager

Dispatcher

System StateUpdate

Poll Events

Update

Update

invoke

Neg. ListNew Job List

Read State

negotiate

job info

invoke

events

Execution of agreement

Neg. Status

New Job List

Server

user

Job Submission

Mal. Job Registration Job

Completion

Event Handler

Coordinator

Node Controllers

Running Applications

Test Rigid Application

OpenSees (Simulation of structural behavior in response to earthquake)

Implemented using PVMMaster-Slave

(coordinator, computing) Input – host list, structure

listEmbarrassingly parallelExecution time

proportional to number of structure

Coordinator

Computing Computing

Host list & list. of structures

H1 H2

Malleable ApplicationCoordinator capable of

negotiationParameters: Min. and

max. processorsAccept offer as long as

total processors between min. and max.

Create and destroy computing process during execution

Expand 2

Accept 1

Negotiation

Coordinator

Computing Computing

RMS

Host 3

H1 H2

Computing

H3

Experimental SetupScheduling Policy

First Come First Serve (FCFS)Pending jobs have priority over running malleable jobsNo job preemptionExpansion and shrinkage candidates are selected in

chronological order of start timeWorkload

Two workloads 40 jobs 120 jobs

Hardware4 processor Pentium Linux cluster connected by 100mb

Ethernet16 processor Pentium Linux cluster connected by

100mb Ethernet

Experimental Results (120 Job Workload)

Job Mix Vs. TAT

1500

1700

1900

2100

0% 20% 40% 60% 80% 100%

% Malleable Job

Avg

. T

AT

Range 2-5

Range 2-6

Range 2-8

Range 4-8

Job Mix Vs. Utilization

0.86

0.88

0.9

0.920.94

0.96

0.98

1

0% 20% 40% 60% 80% 100%

% of Malleable Job

Uti

liza

tio

n

Range 2-5

Range 2-6

Range 2-8

Range 4-8

Avg. Exe. TAT, and Wait time (Range 2-8)

0

500

1000

1500

2000

2500

0% 20% 40% 60% 80% 100%

% of Malleable Job

Tim

e (S

ec.)

Exe. Time

TAT

Wait Time

Max.improvement Utilization – (87 – 100)% Schedule Span – (5058-4390) Avg. TAT – (1969-1657)

With 10% Malleable Utilization – (87-91)% Schedule Span – (5058-4861) Avg. TAT – (1969-1849)

Conclusions From PrototypePresence of malleable jobs in workload

improved performanceHigh percentage of malleable jobs is not

necessary to make significant improvement in performance

On average for malleable jobs, execution time increased, while turn around time decreased

Overhead of managing malleable jobs is low

Modeling Adaptive System with Malleable JobsDeveloped a model for Adaptive System

WorkloadRMSApplications

Numerically simulated the model with synthetic data

Model of Adaptive Applications (ability to negotiate, ability to shrink and expand, cost of adaptation, time to respond, etc)

Model of Negotiationsand Agreement (performance ofcommunication protocol,Convergence etc.)

Model of Scheduling (policies, objectives, decisions, performance etc)

Model of Management (resources, QoS, eventprocessing, etc)

Model of Resource Reallocation(performance, constraints, etc)

Model of Workload (relative composition of evolving/malleable/rigid, characteristic of evolving applications, etc)

Server Scheduler

User(workload)

Negotiator

Running Jobs

Node Controllers

System Status

Dispatcher

6

3

12

Node Controllers

Running Jobs

Server SchedulerUser Negotiator

8 7

11 14

10

13

4

5

One or Multiple Invocation

9

Dispatcher

Interactions between objects1. Submit Job (Job info) 2. Schedule (System State)3 Negotiate (proposal)4 Offer/Counter offer5 Counter offer6 Accept/Reject Notification7 Negotiation outcome (Agreements)8 Execute (schedule)

9. Execute schedule10 . Execute (Agreement)11. Start New Job (Job execution info)12. Launch Job13. Request for negotiation 14. Complete job (Job completion info)

21

Application ModelAssumptions

Applications are parallelComputation is linearly distributed over time. Applications perform negotiation in a non blocking

mannerApplications can adapt at any point during executionOnly processors are considered as resources

Application consists ofOne coordinating process

Negotiation Adaptation

Multiple computing processesApplication consists of phasesDuring a phase

Processors remain unchangedNegotiation occurs

Adaptation results in phase change

Workload ModelAssumption

Workload consists of rigid and malleable applications only

We need realistic workload dataValidating against real rigid dataDevelop realistic malleable workload

Models available for rigid workload Downy (1997), Feitelson(1998), Lubin (2003) etc.

No model available for malleable workloadSelected Downy’s model for modification to

generate malleable workloadValidated against real dataUsed by researcherOpen source code is available

Workload ModelDowny model

Input: No. of jobs, minimum – maximum runtime, minimum – maximum processors

Output: Set of jobs – arrival time, runtime, processor requirement

Extended Downy modelAdditional input: No. of malleable job,

distribution of malleable job, flexibility range

Output: Set of jobs – job type, arrival time, runtime, processor requirement, flexibility range

Negotiation ModelNegotiation cost depends on

Communication time between RMS and applicationRMS response time to an offerApplication response time to an offerNumber of rounds of negotiation

AssumptionsCommunication time and RMS response time is

constantNegotiations is performed sequentially

Negotiation cost variesFrom application to applicationFrom negotiation to negotiation

Negotiation ModelNegotiation cost is approximated by a

single parameter Cn

Cost is stochastically selected from a random ramp distribution.Probability decreases linearly as cost

increases

Adaptation ModelAdaptation cost depends on

Application’s business logicData structure used by the applicationNumber of processors change during

adaptationAdaptation cost varies

From application to applicationFrom adaptation to adaptation

AssumptionsAdaptation cost is linearly proportional to

the change in the number of processorsAdaptation cost doesn’t vary from shrinkage

to expansion

Experimental Design for Investigation of Impact of Parameters on Performance

Each of the parameters were varied while the others kept constantTo isolate the impact of one parameter on

performanceSimulation Data

No of jobs: 1000Runtime: 100 – 3600 secondsFlexibility: 2-128 processors

Parameters

% of Malleable jobs10% -100% in steps of 10

Flexibility of Malleable Jobs2-16, 2-32, 2-64, 2-80, 2-96, 2-112, and 2-128 4-130, 8-134, 12-138, and 16-142

Cost of Negotiation0.0015, 0.003, 0.006, 0.006, 0.0012, 0.012, 0.024,

0.048, 0.96, 0.2, 0.4, 0.8, 2, 4, and 8 seconds. Cost of Adaptation

0.002, 0.004, 0.008, 0.01, 0.02, 0.04, 0.08, 0.2, 0.4, 0.8, 1 second, 2, 4, and 8 seconds

Performance With % of Malleable JobsUtilization as Function of Job Mix

(256 Processors)

0.8

0.85

0.9

0.95

1

0 10 20 30 40 50 60 70 80 90 100

% of Malleable Jobs

Uti

liza

tio

n

Utilization as Function of Job Mix (256 Processors)

0.8

0.85

0.9

0.95

1

0 10 20 30 40 50 60 70 80 90 100

% of Malleable Jobs

Uti

liza

tio

n

Negotiation Cost: 0.0015 Secs.

Adaptation Cost: 0.002 Secs

%Mal Util (256) Util (512)

0 0.84381 0.91165

10 0.92574 0.98121

20 0.99747 0.99186

30 0.9993 0.99112

40 0.99873 0.99447

50 0.9998 0.99178

60 0.9998 0.9986

70 0.99849 0.98803

80 0.99874 0.99327

90 0.99979 0.99488

100 0.99974 0.99629

Avg. TAT With % of Malleable Jobs

%Mal ATAT(256) ATAT (512)

0 109741.53 47580.25

10 101443.34 44222.12

20 93002.55 42997.66

30 92724.88 43387

40 91822.17 42441.01

50 91913.7 42602.09

60 91607.62 42293.82

70 89184.86 40295.11

80 88069.52 38929.32

90 89398.57 38909.31

100 86115.01 38235.4

Negotiation Cost: 0.0015 Secs.

Adaptation Cost: 0.002 Secs

Avg. TAT as Function of Job Mix (1000 Jobs)

80000

84000

88000

92000

96000

100000

104000

108000

0 10 20 30 40 50 60 70 80 90 100

% of Malleable Jobs

Avg

. T

AT

(S

eco

ns)

Avg. TAT as Function of Job Mix (1000 Jobs)

35000

39000

43000

47000

0 10 20 30 40 50 60 70 80 90 100

% of Malleable Jobs

Avg

. T

AT

(S

eco

ns)

Wait, Avg. TAT and Execution Time as function of % of Malleable Jobs

Variation of Avg. Wait, TAT and Execution Time

0

20000

40000

60000

80000

100000

0 10 20 30 40 50 60 70 80 90 100 110

% of Malleable Job

Tim

e (S

ecs.

)

Ave. Wait

Avg. TAT

Aveg. Exec.

% Mal Wait TAT EXEC

0 108791 109741.5 950.56

10 100387.1 101443.3 1056.28

20 90624.57 93002.55 2377.91

30 88932.34 92724.88 3792.46

40 85735.85 91822.17 6086.27

50 84927.36 91913.7 6986.33

60 81548.7 91607.62 10058.87

70 75677.15 89184.86 13507.7

80 72678.14 88069.52 15391.36

90 72811.11 89398.57 16587.46

100 62995.61 86115.01 23119.4

Negotiation Cost: 0.0015 Secs.

Adaptation Cost: 0.002 Secs

Cluster Size: 256 Processors

Summary of Performance with the Variation of Number of Malleable JobsUtilization improves as number of malleable jobs

increasesImprovement saturatesIt is possible achieve maximum utilization with

malleable workload The maximum possible utilization can be achieved

with relatively few malleable jobsAverage turn around time decreases as the

number of malleable jobs increases As the number of malleable jobs increases, the job

execution time increases on average, but the average turn around time decreases. This is because with higher number of malleable jobs, a job has to wait less in the pending queue.

Performance Variation with Lower Bound of Flexibility

Utilization as Function of Minimum Processors (flexibility: 126 processors)

0.89

0.91

0.93

0.95

0.97

0.99

1.01

0 4 8 12 16

Minumum Processors

Uti

lizat

ion

10%

20%

30%

40%

50%

Avg. TAT as function of Minimum Processors (Flexibility: 126 processors

90000

92000

94000

96000

98000

100000

102000

104000

0 4 8 12 16

Minimum Processors

Avg

. T

AT

(se

cs)

10%

20%

30%

40%

50%

Utilization Avg. TATMin. Proc 10 20 30 40 50 10 20 30 40 50

2 0.92574 0.99747 0.9993 0.99873 0.9998 101443.3 93002.55 92724.88 91822.17 91913.74 0.92189 0.99313 0.99962 0.99912 0.99984 101744.2 93733.95 92739.16 92330.06 92231.668 0.91567 0.9827 0.99703 0.99954 0.99912 102379.9 94956.08 93043.13 92599.53 92526.8

12 0.909 0.97298 0.99186 0.99934 0.99466 102688.9 95712.59 93537.43 92742.73 93050.0416 0.90492 0.96574 0.98306 0.99867 0.99313 103083.8 96507.57 94111.55 92804.52 93188.3

Utilization Variation with FlexibilityVariation of Utilization with Number of Malleable Jobs in the

Workload

0.850.870.890.910.930.950.970.991.01

0 10 20 30 40 50 60 70 80 90 100 110

% of Malleable Jobs

Uti

liza

tio

n

14

30

46

62

78

94

110

126

% Mal 2--16 2--32 2--48 2--64 2--80 2--96 2--112 2--12814 30 46 62 78 94 110 126

10 0.86461 0.90117 0.91404 0.91949 0.92277 0.92494 0.92574 0.9257420 0.92978 0.97413 0.98862 0.99095 0.99454 0.99745 0.99683 0.9974730 0.94403 0.97482 0.98663 0.99048 0.99409 0.99677 0.99813 0.999340 0.94996 0.97972 0.98922 0.9941 0.9985 0.99813 0.99824 0.9987350 0.94501 0.97416 0.9877 0.99301 0.99636 0.99802 0.99976 0.999860 0.9504 0.98193 0.9912 0.99749 0.99825 0.99873 0.9994 0.999870 0.94572 0.98252 0.98595 0.98888 0.99312 0.99574 0.99723 0.9984980 0.94963 0.97515 0.98865 0.99431 0.99825 0.99858 0.9984 0.9987490 0.94424 0.97468 0.98505 0.99232 0.99708 0.99821 0.99948 0.99979

100 0.9555 0.98035 0.98943 0.99612 0.99893 0.99856 0.99935 0.99974

Variation of Utilization with Fexibility of Malleable Jobs

0.850.870.890.910.930.950.970.991.01

0 20 40 60 80 100 120 140

Flexibilty

Uti

liza

tio

n

10%

20

30

100%

Avg. TAT Variation with Flexibility

Impact of Flexibility on Avg. TAT for Fixed Minimum processors

84000

88000

92000

96000

100000

104000

0 10 20 30 40 50 60 70 80 90 100 110

% of Malleable Jobs

Uti

liza

tio

n

14

30

46

62

78

Impact of Flexibility on Avg. TAT for Fixed Minimum processor

84000

88000

92000

96000

100000

104000

0 20 40 60 80 100 120 140

Flexibilty (processors)

Avg

. T

AT

(S

ecs)

10%

20

30

100%

% Mal 2--16 2--32 2--48 2--64 2--80 2--96 2--112 2--12814 30 46 62 78 94 110 126

10 102390 101677 101409 101487 101463 101444 101443 10144320 93540 92894 92898 92919 92900 92917 93008 9300230 92640 92701 92699 92698 92717 92726 92719 9272440 91831 91748 91843 91797 91788 91813 91813 9182250 91965 91893 91885 91914 91872 91911 91915 9191360 91400 91590 91423 91385 91458 91442 91553 9160770 88826 89074 89240 89422 89177 89150 89306 8918480 88184 88360 88256 87996 88074 88262 88015 8806990 89063 89088 89546 88863 88814 89248 89224 89398

100 86293 86160 86136 86124 86125 86122 86118 86115

Summary of Performance with the Variation of FlexibilityPerformance decreases as the lower bound of

flexibility increasesThe impact of lower bound decreases as the

number of malleable job increases For a given flexibility utilization increases

with the number of malleable jobs and saturates

At saturation point, utilization increases as flexibility increases and saturates at certain flexibility

Variation of Utilization with Negotiation Cost

Neg Cost Utilization10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

0.0015 0.92574 0.99747 0.9993 0.99873 0.9998 0.9998 0.99849 0.99874 0.99979 0.999740.003 0.92574 0.99739 0.9993 0.99873 0.99979 0.99879 0.99849 0.99873 0.99979 0.999740.006 0.92573 0.99738 0.99928 0.99949 0.99978 0.99983 0.999848 0.99873 0.99978 0.999740.012 0.92573 0.99746 0.99926 0.99869 0.99976 0.99978 0.99846 0.99871 0.99977 0.999740.024 0.92574 0.99739 0.99927 0.99866 0.99973 0.99974 0.99842 0.99868 0.99976 0.999740.048 0.92571 0.99738 0.99923 0.99849 0.99964 0.99968 0.99817 0.99864 0.9993 0.999730.096 0.92569 0.99725 0.99913 0.99862 0.99948 0.99948 0.9982 0.99537 0.99924 0.99972

0.2 0.92564 0.99704 0.99866 0.99662 0.99866 0.99874 0.99758 0.99825 0.99951 0.99970.4 0.92554 0.996 0.99697 0.99679 0.99729 0.99669 0.9968 0.997 0.99903 0.999650.8 0.92512 0.99553 0.99642 0.9929 0.99527 0.99544 0.99472 0.99074 0.99729 0.99956

2 0.92433 0.99307 0.99279 0.98726 0.98848 0.98816 0.9877 0.99096 0.99528 0.999234 0.9221 0.9864 0.9816 0.97445 0.96888 0.97615 0.97734 0.98686 0.98974 0.998758 0.92051 0.97961 0.96659 0.94554 0.94841 0.94637 0.95367 0.96609 0.96609 0.9967

Utilization as Function of Job Mix

0.9

0.92

0.94

0.96

0.98

1

1.02

0 10 20 30 40 50 60 70 80 90 100

% of Malleable Jobs

Uti

lizat

ion

8 Secs.

4 Secs

2 Secs.

.0015 Secs

0.8 Secs

Utilization as function of Negotiation Cost

0.9

0.92

0.94

0.96

0.98

1

1.02

0 1 2 3 4 5 6 7 8 9

Negotiation Cost (Seconds)

Uti

liza

tio

n

10% 20% 60% 100%

Variation of Avg. TAT with Negotiation Cost

Avg. TAT as Function of Job Mix

8400086000880009000092000940009600098000

100000102000104000

0 10 20 30 40 50 60 70 80 90 100

% of Malleable Jobs

Avg

. T

AT

(S

ecs)

1.5 ms

2 Sec.

4 Sec.

8 Sec.

0.8 Secs

Avg. TAT as Function of Negotiation Cost

85000

90000

95000

100000

105000

0 1 2 3 4 5 6 7 8 9

Negotiation Cost (Secs.)

Avg

. T

AT

(S

ecs.

)

10%

20%

60%

100%

Neg Cost Average Turn Around Time10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

0.0015 101443 93002 92724 91822 91913 91607 89184 88069 89398 861150.003 101443 93003 92725 91822 91914 91608 89185 88070 89398 861150.006 101443 93003 92725 91817 91915 91554 89186 88071 89399 861150.012 101444 93005 92726 91843 91917 91590 89187 88073 89399 861150.024 101445 93006 92742 91829 91918 91592 89192 88077 89401 861150.048 101446 93007 92747 91852 91913 91523 89345 88090 89327 861150.096 101447 92977 92740 91829 91959 91567 89385 88144 89332 86115

0.2 101452 93061 92758 92078 92060 91549 89384 88424 89420 861150.4 101461 93083 92939 91933 92117 91827 89448 88372 89363 861160.8 101485 93156 92856 92477 92342 91915 89595 88570 89396 86117

2 101581 93367 93278 92846 93026 92776 90359 88679 90057 861594 101710 93921 94559 94777 94821 94181 90910 89306 90387 861448 101902 94691 95637 97250 97339 96759 93732 92122 90857 86374

Variation of Number of Negotiation with Variation of Number of Malleable Jobs

% Mal No.of Neg.0 0

10 59920 169930 275040 370950 407260 609070 610680 463690 2882

100 306

Variation of Negotiation with Number of Malleable Jobs

0

1000

2000

3000

4000

5000

6000

7000

0 10 20 30 40 50 60 70 80 90 100

% of Malleable JobsN

o.

of

Neg

oti

atio

ns

Summary of Performance with the Variation of Negotiation Cost

Negotiation cost up to 0.8 second has no significant impact on performance

Negotiation cost doesn’t impact 10% and 100% job mixes

For the same negotiation cost as the number of malleable job increases the utilization decreases, and then the utilization increases as the number of malleable increases further

Variation of Utilization with Adaptation Cost

Impact of Adaptation Cost

0.8

0.85

0.9

0.95

1

0 10 20 30 40 50 60 70 80 90 100

% of Malleable Jobs

Uti

liza

tio

n

2 ms.

0.08 Sec.

1 Secs.

2 Secs.

4 Secs.

8 Secs.

Impact of Adaptation Cost

0.88

0.9

0.92

0.94

0.96

0.98

1

1.02

0 1 2 3 4 5 6 7 8

Adaptation Cost (Secs.)

Uti

liza

tio

n

10%

20%

30%

40%

50%

Adaptation UtilizationCost 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

0.002 0.92574 0.99747 0.9993 0.99873 0.9998 0.9998 0.99849 0.99874 0.99979 0.999740.004 0.92574 0.99733 0.99907 0.99894 0.99973 0.99974 0.99844 0.99879 0.99973 0.999710.008 0.92574 0.99737 0.99882 0.99874 0.99961 0.99963 0.99852 0.99863 0.9996 0.999640.01 0.92574 0.99727 0.99867 0.99867 0.9995 0.99958 0.99861 0.99847 0.99954 0.999610.02 0.92574 0.99645 0.99879 0.99894 0.99923 0.99931 0.99797 0.99816 0.9989 0.999450.04 0.92568 0.99618 0.99829 0.99801 0.99858 0.99882 0.99753 0.99769 0.99822 0.999130.08 0.9267 0.99544 0.99737 0.99652 0.99747 0.99736 0.99668 0.99588 0.99708 0.998490.2 0.92542 0.9899 0.99263 0.99259 0.99387 0.99505 0.99346 0.99192 0.99353 0.996580.4 0.92512 0.98514 0.98681 0.98795 0.98917 0.98933 0.98866 0.98433 0.98802 0.993280.8 0.9233 0.97772 0.97528 0.97671 0.97995 0.97863 0.98058 0.97214 0.97787 0.98677

1 0.92256 0.97392 0.97156 0.97412 0.97773 0.97598 0.97581 0.96689 0.97184 0.984362 0.91823 0.95577 0.95832 0.94937 0.95843 0.95465 0.95107 0.94149 0.94814 0.970864 0.91612 0.94205 0.93835 0.92872 0.93403 0.9289 0.92023 0.89598 0.90425 0.947318 0.9032 0.90535 0.90638 0.89453 0.90023 0.88113 0.87146 0.84255 0.86391 0.91249

Variation of Avg. TAT with Adaptation Cost

Impact of Adaptation Cost

84000

89000

94000

99000

0 10 20 30 40 50 60 70 80 90 100

% of Malleable Jobs

Avg

. T

AT

(S

ecs.

)

2 ms.

0.4 Secs.

1 Secs.

2 Secs.

4 Secs.

8 Secs.

Impact of Adaptation Cost on Avg. TAT

91000

93000

95000

97000

99000

101000

0 1 2 3 4 5 6 7 8

Adaptation Cost (Secs.)

Avg

. T

AT

(Sec

s.)

10%20%30%40%50%

Adaptation TATCost 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

0.002 101443 93002 92724 91822 91913 91607 89184 88069 89398 861150.004 101443 93008 92741 91815 91918 91609 89187 88226 89402 861150.008 101443 92995 92696 91817 91874 91612 89261 88109 89408 861160.01 101443 93003 92756 91820 91968 91614 89008 88025 89411 861170.02 101444 93017 92733 91859 91946 91540 89045 88244 89288 861190.04 101452 93069 92828 91897 92019 91577 89265 88399 89300 861230.08 101456 93134 92865 91959 92097 91567 89400 88376 89259 861330.2 101475 93610 93197 92211 92279 91932 89493 88489 89670 861590.4 101476 94007 93668 92605 92591 92057 89797 88964 90222 862000.8 101391 94545 94761 93129 93081 92767 90145 89974 90609 86259

1 101397 94722 94954 93262 93149 92852 90308 89866 90560 862672 101566 96088 95352 94754 94143 93787 91494 91340 92057 863784 101557 96500 96069 95469 95430 94421 92818 93935 94133 865568 101793 97737 97504 96220 96064 95308 94386 95679 95812 86788

Summary of Performance with the Variation of Adaptation Cost

Performance decreases as adaptations cost increases

Adaptation cost doesn’t impact 10% and 100% job mixes significantly

The impact of adaptation cost on performance more pronounced compared to the impact of negotiation cost

Conclusions From SimulationPerformance improves as number of malleable jobs

increases and it is possible to achieve maximum utilization with relatively few malleable jobs

Performance decreases as the lower bound of flexibility increases

The impact of lower bound decreases as the number of malleable job increases

For a given flexibility, utilization increases with the number of malleable jobs and saturates

Negotiation cost does not have significant impact on performance

Adaptation cost impact performance significantlyNumber of negotiation varies with the number of malleable

applications in the workload

Future WorkScheduling algorithm

Policy regarding selection of negotiation candidatesFailed Negotiation

Further investigation of impact of flexibilityLate release of resourcesModel involving evolving applicationsProgramming model for adaptive applicationsExtending the present work in Grid

environment

AcknowledgementDr. Tomasz HauptDr. Ioana BanicescuMahbubur Rashid and Satya GosulaCenter for Advanced Vehicular Systems at

Mississippi State UniversityDepartment of Computer Science and

Engineering at Mississippi State UniversityDepartment of Computer Science at Tennessee

Technological UniversityNSF Grant EIA 0103594

References1. D. G. Feitelson and L. Rudolph, “Towards convergence in job scheduling for parallel

super computers,” in Job Scheduling Strategies for Parallel Processing, Vol. 1162, Lecture Notes in Computer Science D. G. Feitelson and L. Rudolph Eds. Springer-Verilag, 1996, pp 1-26.

2. S.S. Vadhiyar and J. Dongarra, “SRS: A framework for developing malleable and migratable parallel applications for distributed systems”, Parallel Processing Letters, Vol. 13, No. 2 (2003) 291-312

3. J. Hungershofer, “On the Combined Scheduling of Malleable and Rigid Jobs”, in Proceedings of the 16th Symposium on Computer Architecture and High Performance Computing, 2004.

4. J. Hungershofer, A. Streit, and Jens-Michael Wierum. Efficient Resource Management for Malleable Applications. Technical Report PC2, TR-003-01, December 2001, http://wwwcs.upb.de/pc2/papers/files/394.pdf

5. J. E. Moreira and V. K. Naik. Dynamic Resource Management on Distributed Systems Using Reconfigurable Applications. IBM Journal of Research and Development, Vol. 41, No. 3, May 1997, pp 303 – 330

6. R. Jha, M. Muhammad, S. Yalamanchili, K. Schwan, D. Ivan Rosu, and C. de Castro, "Adaptive resource allocation for embedded parallel applications", in Proceedings of the 3rd International Conference on High Performance Computing", Trivandrum India, December 1996.

7. D. Ivan Rosu, K. Schwan, S. Yalamanchili, and R. Jha, "On adaptive resource allocation for complex real-time applications", in Proceedings of the 18th IEEE Real-Time Systems Symposium, San Francisco, December 1997.

References8. Allen B. Downey, ``A Parallel Workload Model and Its

Implications for Processor Allocation''. 6th Intl. Symp. High Performance Distributed Comput., Aug 1997.

9. Uri Lublin and Dror G. Feitelson, The Workload on Parallel Supercomputers: Modeling the Characteristics of Rigid Jobs. J. Parallel & Distributed Comput. 63(11), pp. 1105-1122, Nov 2003.

10. D. G. Feitelson, ``Packing schemes for gang scheduling”. In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (Eds.), Springer-Verlag, 1996, Lect. Notes Comput. Sci. vol. 1162, pp. 89-110.

11. Open System for Earthquake Engineering Simulations, http://opensees.Berkeley.edu

QuestionsContact

sghafoor@tntech.edu