X

Download Systematic Concurrency Testing using CHESS PowerPoint Presentation


Login   OR  Register
X


Iframe embed code :



Presentation url :

X

Description :

Available Systematic Concurrency Testing using CHESS powerpoint presentation for free download which is uploaded by search an active user in belonging ppt presentation Sports & Recreation category.

Tags :

Systematic Concurrency Testing using CHESS

Home / Sports & Recreation / Sports & Recreation Presentations / Systematic Concurrency Testing using CHESS PowerPoint Presentation

Systematic Concurrency Testing using CHESS PowerPoint Presentation

Ppt Presentation Embed Code   Zoom Ppt Presentation

About This Presentation


Description : Available Systematic Concurrency Testing using CHESS powerpoint presentation for free download which... Read More

Tags : Systematic Concurrency Testing using CHESS

Published on : Feb 10, 2014
Views : 404 | Downloads : 0


Download Now

Share on Social Media

             

User Presentation
SlidesFinder: Advertise with us
Related Presentation
Free PowerPoint Templates
Donuts Free PowerPoint Template

Donuts

Views : 611

Meditation Free PowerPoint Template

Meditation

Views : 212

Injecting Free PowerPoint Template

Injecting

Views : 260

Email Free PowerPoint Template

Email

Views : 276

Obesity Free PowerPoint Template

Obesity

Views : 647

Happy Birthday Free PowerPoint Template

Happy Birthday

Views : 384

Heart Free PowerPoint Template

Heart

Views : 280

Nature Abstract Free PowerPoint Template

Nature Abstract

Views : 274

Agriculture Free PowerPoint Template

Agriculture

Views : 643

Butterfly Free PowerPoint Template

Butterfly

Views : 390

Synapse Free PowerPoint Template

Synapse

Views : 515

Laboratory Free PowerPoint Template

Laboratory

Views : 457

Airplane Free PowerPoint Template

Airplane

Views : 217

Dollar Free PowerPoint Template

Dollar

Views : 265

Herbal Pills Free PowerPoint Template

Herbal Pills

Views : 377

Dental Brush Free PowerPoint Template

Dental Brush

Views : 301

Bales Free PowerPoint Template

Bales

Views : 363

Quit Smoking Free PowerPoint Template

Quit Smoking

Views : 414

Slide 1 - CHESS: Systematic Concurrency Testing Tom Ball, Sebastian Burckhardt, Madan Musuvathi, Shaz Qadeer Microsoft Research http://research.microsoft.com/CHESS/
Slide 2 - Testing concurrent programs is HARD Rare thread interleavings expose bugs Coverage problem Testing misses thread interleavings that expose errors Reproducibility problem Concurrency bugs == Heisenbugs Not reproducible  hard to debug Crash dumps don’t help
Slide 3 - Thread interleavings x++; x++; x*=2; x*=2; 4 1 0 2 8 4 3 2 5 6 Thread 1 Thread 2 1 0 0 2 2 2 1 4 3
Slide 4 - Concurrency testing today Concurrency testing == stress testing Example: testing a concurrent queue Create 100 threads performing queue operations Run for days/weeks Stress increases the interleaving variety, but Not systematic: might miss interleavings Not predictable: cannot find the same error again Makes any error found hard to debug
Slide 5 - 1 Why stress is not sufficient
Slide 6 - Concurrency testing : what we need Methodology and tools to systematically and predictably test thread interleavings
Slide 7 - CHESS in a nutshell Replace the OS scheduler with a demonic scheduler Systematically explore all scheduling choices ConcurrentProgram Win32 API Kernel Scheduler Demonic Scheduler
Slide 8 - CHESS will run this program 6 times exploring all the different interleavings x++; x++; x*=2; x*=2; 4 1 0 2 8 4 3 2 5 6 Thread 1 Thread 2 1 0 0 2 2 2 1 4 3
Slide 9 - 2 Don’t stress, use CHESS
Slide 10 - CHESS architecture Kernel: Threads, Scheduler, Synchronization Objects While(not done) { TestScenario() } TestScenario() { … } Program CHESS CHESS runs the scenario in a loop Every run takes a different interleaving Every run is repeatable Win32 API Intercept synch. & threading calls To control and introduce nondeterminism Detect Assertion violations Deadlocks Dataraces Livelocks
Slide 11 - CHESS methodology generalizes Need wrappers for every concurrency API CHESS has wrappers for Win32, .NET, Singularity Wrappers understand the semantics of the API Expose nondeterminism in the API Looking for volunteers to build wrappers for Linux and Java
Slide 12 - CHESS clients PCP = Parallel Computing Platform (for multi/many-cores) PLINQ: Parallel LINQ CDS: Concurrent Data Structures STM: Software Transactional Memory TPL: Task Parallel Library ConcRT: Concurrency RunTime CCR: Concurrency Coordination Runtime Dryad Part of COSMOS Singularity/Midori CHESS can systematically test the boot and shutdown process
Slide 13 - Stateless model checking [Verisoft ‘97] Systematically enumerate all paths in a state-space graph Don’t capture program states Capturing states is extremely hard for large programs Effective for message-passing programs CHESS applies stateless model checking for shared-memory multithreaded programs
Slide 14 - Outline Preemption bounding [PLDI ‘07] Fair stateless model checking [PLDI ‘08] Sober [CAV ’08, EC2 ‘08] FeatherLite Concurrency Explorer [EC2 ‘08]
Slide 15 - Outline Preemption bounding Makes CHESS effective on deep state spaces Fair stateless model checking Sober FeatherLite Concurrency Explorer
Slide 16 - x = 1; … … … … … y = k; State space explosion Thread 1 Thread n x = 1; … … … … … y = k; … n threads k steps each Number of executions = O( nnk ) Exponential in both n and k Typically: n < 10 k > 100 Limits scalability to large programs Goal: Scale CHESS to large programs (large k)
Slide 17 - x = 1; if (p != 0) { x = p->f; } Preemption bounding Prioritize executions with small number of preemptions Two kinds of context switches: Preemptions – forced by the scheduler e.g. Time-slice expiration Non-preemptions – a thread voluntarily yields e.g. Blocking on an unavailable lock, thread end x = p->f; } x = 1; if (p != 0) { p = 0; Thread 1 Thread 2 preemption non-preemption
Slide 18 - Polynomial state space Terminating program with fixed inputs and deterministic threads n threads, k steps each, c preemptions Number of executions <= nkCc . (n+c)! = O( (n2k)c. n! ) Exponential in n and c, but not in k x = 1; … … … … … y = k; x = 1; … … … … … y = k; Thread 1 Thread 2 x = 1; … … … … x = 1; … … … … y = k; … … y = k; Choose c preemption points Permute n+c atomic blocks
Slide 19 - 3 Preemption bounding
Slide 20 - Find lots of bugs with 2 preemptions Acknowledgement: testers from PCP team
Slide 21 - So, is CHESS is unsound? Soundness: prove that the program is correct for a given input test harness Need to exhaustively explore all interleavings For small programs, CHESS is sound Iteratively increase the preemption bound Preemption bounding helps scale to large programs A good “knob” to trade resources for coverage Better search algorithms  more coverage faster Partial-order reduction Modular testing of loosely-coupled programs
Slide 22 - Outline Preemption bounding Makes CHESS effective on deep state spaces Fair stateless model checking Makes CHESS effective on cyclic state spaces Enables CHESS to find liveness violations (livelocks) Sober FeatherLite Concurrency Explorer
Slide 23 - Concurrent programs have cyclic state spaces Spinlocks Non-blocking algorithms Implementations of synchronization primitives Periodic timers … L1: while( ! done) { L2: Sleep(); } M1: done = 1; Thread 1 Thread 2 ! done L2 ! done L1 done L2 done L1
Slide 24 - A demonic scheduler unrolls any cycle ad-infinitum ! done done ! done done ! done done while( ! done) { Sleep(); } done = 1; Thread 1 Thread 2 ! done
Slide 25 - Depth bounding ! done done ! done done ! done done ! done Prune executions beyond a bounded number of steps Depth bound
Slide 26 - Problem 1: Ineffective state coverage ! done ! done ! done ! done Bound has to be large enough to reach the deepest bug Typically, greater than 100 synchronization operations Every unrolling of a cycle redundantly explores reachable state space Depth bound
Slide 27 - Problem 2: Cannot find livelocks Livelocks : lack of progress in a program temp = done; while( ! temp) { Sleep(); } done = 1; Thread 1 Thread 2
Slide 28 - Key idea This test terminates only when the scheduler is fair Fairness is assumed by programmers All cycles in correct programs are unfair A fair cycle is a livelock while( ! done) { Sleep(); } done = 1; Thread 1 Thread 2 ! done ! done done done
Slide 29 - We need a fair demonic scheduler Avoid unrolling unfair cycles Effective state coverage Detect fair cycles Find livelocks (violations of fair termination) ConcurrentProgram Test Harness Win32 API Demonic Scheduler Fair Demonic Scheduler
Slide 30 - Fair termination allows CHESS to check for arbitrary liveness properties Example: Good Samaritan assumption Forall threads t : GF scheduled(t)  GF yield(t) A thread when scheduled infinitely often yields the processor infinitely often Examples of yield: Sleep(), ScheduleThread(), asm {rep nop;} Thread completion while( ! done) { Sleep(); } done = 1; Thread 1 Thread 2
Slide 31 - Outline Preemption bounding Makes CHESS effective on deep state spaces Fair stateless model checking Makes CHESS effective on cyclic state spaces Enables CHESS to find liveness violations (livelocks) Sober Detect relaxed-memory model errors Do not miss behaviors only possible in a relaxed memory model FeatherLite Concurrency Explorer
Slide 32 - C# Example volatile bool isIdling; volatile bool hasWork; //Consumer thread void BlockOnIdle(){ lock (condVariable){ isIdling = true; if (!hasWork) Monitor.Wait(condVariable); isIdling = false; } } //Producer thread void NotifyPotentialWork(){ hasWork = true; if (isIdling) lock (condVariable) { Monitor.Pulse(condVariable); } } 32
Slide 33 - Key pieces of code on previous slide: On x86, hardware may perform store late Bug: Producer thread does not notice waiting Consumer, does not send signal Store ii, 1 Example: Store Buffer Vulnerability Store ii, 1 volatile int ii = 0; volatile int hw = 0; Load hw, 0 Load ii, 1 Store hw, 1 Consumer Producer 0 33
Slide 34 - Sober algorithm Programmers assume sequential-consistency (SC) Insert synchronizations & fences to counter memory-model relaxations Sober checks if a program is memory-model safe i.e., program has only SC executions in a memory model Reports any such violation as an error Sober is a dynamic monitor that checks if any SC execution can be extended to a non-SC execution Theorem: CHESS + Sober guarantees memory-model safety
Slide 35 - Outline Preemption bounding Makes CHESS effective on deep state spaces Fair stateless model checking Makes CHESS effective on cyclic state spaces Enables CHESS to find liveness violations (livelocks) Sober Detect relaxed-memory model errors Do not miss behaviors only possible in a relaxed memory model FeatherLite A light-weight data-race detection engine (<20% overhead) Concurrency Explorer
Slide 36 - Outline Preemption bounding Makes CHESS effective on deep state spaces Fair stateless model checking Makes CHESS effective on cyclic state spaces Enables CHESS to find liveness violations (livelocks) Sober Detect relaxed-memory model errors Do not miss behaviors only possible in a relaxed memory model FeatherLite A light-weight data-race detection engine (<20% overhead) Concurrency Explorer First-class concurrency debugging
Slide 37 - Conclusion Don’t stress, use CHESS CHESS binary and papers available at http://research.microsoft.com/CHESS Stateless model checking is very effective Preemption bounding to scale to deep state spaces Fair demonic scheduler to handle nonterminating programs Need better testing and debugging methodologies for concurrent programs
Slide 38 - Questions