c laus b rabrand © c oncurrency (q3,’07) f eb 05, 2007 c oncurrency claus brabrand...

Post on 18-Dec-2015

215 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

CLAUS BRABRAND © CONCURRENCY (Q3,’07) FEB 05, 2007

CONCURRENCY

Claus Brabrandbrabrand@daimi.au.dk

[ http://www.daimi.au.dk/~brabrand/ ]

CLAUS BRABRAND © CONCURRENCY (Q3,’07) FEB 05, 2007

WELCOME !

CLAUS BRABRAND[ 3 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Agenda (Course Presentation)

Course Presentation: Weekly schedule Aims & goals Exam & assessment Roles: students & teachers Motivation: ”Why bother learning about Concurrency!?!” Materials (book & notes)

Week 1: ”Processes and Threads” Concepts Models (FSP) Practice (Java) Models (FSP) (Java) Practice

CLAUS BRABRAND[ 4 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Weekly Schedule

Lectures: Mondays 11:15-14:00 (sometimes a bit less)

Theoretical Exercise Classes (aka. “TØ”): Tuesdays 08:15-10:00 [ shannon.164 ]

Practical Programming Labs (aka. “Lab”): Thursdays 10:15-12:00 [ zuse.127 ]

Studying: Exercises, Reading, and Thinking(!)

Compulsory homework: Hand-ins (Sunday deadlines), Mini-project

3 hrs/week

2 hrs/week

2 hrs/week

? hrs/week

? hrs/week

Expected total: 15 hrs/week

CLAUS BRABRAND[ 5 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Course Aims & Goals

YourAims & Goals:

at huske udenad… at identificere brug af concurrency i (Java) programmer; at beskrive concurrency mekanismer i programmeringssprog (Java); at forklare fundamentale problemer, teknikker og løsninger indenfor concurrency; at forklare sammenhængen mellem modeller og programmer generelt; at relatere konkrete modeller og programmer; at sammenligne konkrete modeller; at ræsonnere om egenskaber af konkrete modeller og programmer; at implementere konkrete modeller i programmeringssprog (Java); at anvende ovenstående med henblik på at skabe konkrete modeller af systemer.

CLAUS BRABRAND[ 6 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Exam Eksamenerne er konstrueret til at evaluere din dybdeforståelse af

concurrency i overensstemmelse med dine mål (se ovenfor).

Jo bedre du kan ovenstående mål, jo bedre vil du klare dig til eksamen; du vil således kunne bruge målene overfor dig selv som en checkliste for din egen forståelse. Som en direkte konsekvens heraf vil der ikke være hverken "type"- eller "huske"-opgaver (som ikke nødvendigvis forudsætter dybdeforståelse).

Eksamen: 13-skala karakter på baggrund af: (gruppe) mini-projektet samt (individuel) multiple-choice prøve

...som begge vil være centreret omkring ovenstående mål

NB: For at blive indstillet til eksamen, skal alle afleveringsopgaver være afleveret - og godkendt.

CLAUS BRABRAND[ 7 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Concurrency: Aims & Goals Exam

Motivation Roles Program Project Exam Materials Classes Webboard

Homepage [ http://www.daimi.au.dk/~brabrand/concurrency/ ]

CLAUS BRABRAND © CONCURRENCY (Q3,’07) FEB 05, 2007

YOUR ROLE

Active Learning (!)

CLAUS BRABRAND[ 9 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Consider the following “alphabet” (60 secs):

Exercise: Now write my office phone#: 89425771 ?

Exercise 1

1. 2. 3. 4. 5.

6. 7. 8. 9. 0.

CLAUS BRABRAND[ 10 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Exercise 1 (cont’d)

Suppose I now showed you…:

Exercise: Now write my office phone#: 89425771 ?

1 2 3

4 5 6

7 8 9

CLAUS BRABRAND[ 11 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Exercise 1 (cont’d)

So what is the point? Random information is really hard to remember We learn (efficiently) by associating (building)

new unknown information / with (on) old known information

Knowledge is constructed as a result of: Learner’s activity (!)

In short: “You need to study”: i.e. read, make exercises, and think(!)

CLAUS BRABRAND © CONCURRENCY (Q3,’07) FEB 05, 2007

COLLABORATION

Teaching / Learning Collaboration

CLAUS BRABRAND[ 13 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Roles (Teaching/Learning Collaboration)

During learning period (i.e., before the exam): Teaching/Learning collaboration:

You = “Learner” Learning responsibility

Me = “Teacher” (and Teaching Assistant) (To the best of my ability) aid you in learning

During the exam period: You = “Performer”

Demonstrate what you have learned

Me = “Evaluator” (Neutrally) assess your level of understanding

CLAUS BRABRAND © CONCURRENCY (Q3,’07) FEB 05, 2007

MOTIVATION

“Why bother learning about Concurrency?!?”

CLAUS BRABRAND[ 15 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

What is a Concurrent Program?

A sequential program has a single thread of control

A concurrent program has multiple threads of control:

Perform multiple computations in parallel Control multiple external activities which

occur at the same time

CLAUS BRABRAND[ 16 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Why Concurrent Programming? More appropriate program structure

Concurrency reflected in program

Performance gain from multiprocessing HW Parallelism

Increased application throughput An I/O call need only block one thread

Increased application responsiveness High-priority thread for user requests

CLAUS BRABRAND[ 17 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Concurrency vs. Sequentiality

Sequential programming: Inherently deterministic Describe computation as a “reduction” of expressions to values Termination often desirable Resulting value is of primary interest and focus

Concurrent programming: Inherently non-deterministic Describe execution as “process evolution” Describe possible executions (aka. execution traces) Describe possible interactions during execution Describe interaction with an environment Non-termination often desirable (e.g. Op.Sys., Control sys, Cell-phone, …) Resulting “value” is not (necessarily) interesting

CLAUS BRABRAND[ 18 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Concurrency is much Harder

Harder than sequential programming: Huge number of possible executions Inherently non-deterministic Parallelism conceptually harder

Consequences: Programs are harder to write(!) Programs are harder to debug(!) Errors are not always reproducible(!) New kinds of errors possible(!):

Deadlock, starvation, priority inversion, interference, …

CLAUS BRABRAND[ 19 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Example Concurrency Problems

Therac-25 Radiation Therapy ’85-’87

Massive overdoses (6 deaths / amputations)!

Mars Pathfinder July ’97

Periodic resets

Win95/98 w/ 3rd-Party Device Drivers late ’90es

Dysfunction (“blue screen of death”)!

...on mars!

CLAUS BRABRAND[ 20 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Concurrency Problems (cont’d)

Mobile Phones ’00-…

Freeze and odd behaviors (really annoying)!

Cruise Control System Model ’86 [Grady Booch]

Accellerated after car ignition (car crashes)!

CLAUS BRABRAND[ 21 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

…and what about?

Air Plane Control System

Dysfunction (plane crash)!

Nuclear Powerplant Control System

Core melt-down (“China-syndrome”)!

CLAUS BRABRAND[ 22 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Problem: System Development?

In the presence of all these errors: deadlock, starvation, priority inversion, interference,

anti-cooperation, un-intended execution traces, un-fairness, …

How to…: 1. …design a system that “works” ? 2. …verify that the system is “safe” ? 3. …verify that the system “meets its specification” ?

…and: What does “works”, “safe”, and “to meet a specification” mean ?!?

CLAUS BRABRAND[ 23 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Solution: Modeling

“Models come to the rescue”: Create models (~ architecture, bridge construction, …)

Formal modeling (e.g., “FSP”) permits: (Offline) Reasoning understanding (Runtime) Testing confidence (C-time) Property Verification safety (C-time) Specification Verification correctness

Note: “Errors are much cheaper to commit in models than in implementations”

“Never send a human to do a machine’s job”-- A.Smith (’99)

auto-mate

CLAUS BRABRAND[ 24 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Methodology: Model-based Design

Design abstract model

Decompose model

Reason/Test/Verify model individual parts and whole

Recompose insights make model safe

Impl. concrete program

concretize

REAL PROBLEM

SAFE MODEL

SAFE PROGRAM

abstract

? ?

?

?

MODEL

testreason

verify

CLAUS BRABRAND[ 25 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Program worldModel world

ConcreteAbstract

P Mabstraction

concretization

Program vs. Model World

(Java) (FSP)

What discerns a really good programmer from one that is not so good is the capability of moving (consciously or unconsciously) between the concrete world of programs and the abstract world of semantic models (via abstraction and concretization).

Specifically, such a programmer is capable of (consciously or unconsciously): - 1) abstracting programs into models - 2) reasoning about the models - 3) concretizing the insights back into the world of programs

CLAUS BRABRAND[ 26 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Program worldModel world

ConcreteAbstract

P M

1. P ? 3. M ?

4. validate

5. M !7. P !

2. abstract

6. concretize

Program / Model Verification

(Java) (FSP)

CLAUS BRABRAND[ 27 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Program worldModel world

ConcreteAbstract

~

P

P’

M

M’

1. P ~ P’ ?2. abstract

3. M ~ M’ ?

4. relate

5. M ~ M’ !6. concretize7. P ~ P’ !

Program / Model Relationship

(Java) (FSP)

CLAUS BRABRAND[ 28 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Glasses of Abstraction…

It’s Amazing…

Abstract away computational details: see only concurrency aspects

at appropriate level of abstraction!

Only$12,95

“See only the relevant!”

“The Glasses of Abstraction”

COMPLEX PROGRAMCOMPLEX PROGRAM

MODELMODEL

abstraction

CLAUS BRABRAND[ 29 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Modeling: Level of Abstraction

Consider a client/server system:

Extremely abstract (high level of abstraction):

Appropriate (level of abstraction) for … :

Extremely concrete (low level of abstraction):

System = (event -> System).

NAND_Gate = ...Transistor = ...Accumulator = ......Client = ...

Server = (request->compute->reply->Server).Client = (calc->request->wait->reply->Client).Database = ...

high

er le

vel o

f ab

stra

ctio

n

CLAUS BRABRAND[ 30 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Notes on Skill Acquisition

Your skill acquisition will progress according to the following stages of learning:

4. Unconscious competence 3. Conscious competence 2. Conscious incompetence 1. Unconscious incompetence

Course aim (and measurement): Level 3 Later in life, you probably mostly need level 4

…but you need to get there via level 3

CLAUS BRABRAND[ 31 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Modeling Language (e.g. “FSP”)

Compositional: ||

Break things into (several) smaller things

Algebraic: ,

Intuitive ideal and eases automated verification

Syntactic:

Provide basis for programming languages

P || Q P Q

big

P || Q Q || P P || (Q || R)

USER =def (acquire -> use -> release -> USER).

(P || Q) || R

CLAUS BRABRAND[ 32 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Parallel- vs. Concurrent Programming

Strategy: Optimal strategy for a particular goal Use available resources efficiently

Safety: Conceptually independent players Control interaction and “rules”

The Football Match Analogy:“An analogy that one can make is with football*; - the coach of the team is a parallel programmer while - the referee is a concurrent programmer” -- [P.Panangaden, ’96]

The Referee (~ the concurrent programmer):-- Make sure what is happening is a soccer match

The Trainer (~ the parallel programmer):-- Make sure my agents are performing “optimally”

*/ interpret appropriately on either side of the Atlantic

CLAUS BRABRAND[ 33 ]

© CONCURRENCY (Q3,’07) FEB 05, 2007

Course Aims & Goals

YourAims & Goals:

at huske udenad… at identificere brug af concurrency i (Java) programmer; at beskrive concurrency mekanismer i programmeringssprog (Java); at forklare fundamentale problemer, teknikker og løsninger indenfor concurrency; at forklare sammenhængen mellem modeller og programmer generelt; at relatere konkrete modeller og programmer; at sammenligne konkrete modeller; at ræsonnere om egenskaber af konkrete modeller og programmer; at implementere konkrete modeller i programmeringssprog (Java); at anvende ovenstående med henblik på at skabe konkrete modeller af systemer.

T

top related