Hazim Shafi Principal Architect Microsoft Corporation TL19.

Microsoft Visual Studio:Bringing Out The Best In Multicore Systems

Hazim ShafiPrincipal ArchitectMicrosoft Corporation

TL19

Motivation Parallel application development

The basics The role of tools

Exploiting parallelism Current and future programming models

Parallel performance tuning New Visual Studio tools

Demos throughout

Outline

Visual Studio 2010 includes many parallel programming features Reduce programming complexity

Managed and native Debugging support

Concurrency analysis tools Causes of inefficiency Opportunities for parallelism

Want more details? References to related PDC talks and symposia

Outline

Multicore processors are here Single thread performance is ~flat Performance burden shifted to software Enables new user experiences

Major concern for Windows developers Performance suffers with new features Need help parallelizing applications

Parallel Computing Platform team Ease of developing, debugging, and tuning

parallel applications Help is on its way in Visual Studio 2010

Motivation

Visual Studio 2010Tools/programming models/runtimes

Parallel Pattern library

Resource manager

Task scheduler

Task Parallel library

PLINQ

Managed library Native libraryKey:

Threads

Operating system

Concurrency runtime

Programming models

Agentslibrary

Thread pool

Task scheduler

Resource manager

Data structures D

ata

stru

ctur

es

Tools

Tools

ParallelDebugger

Toolwindows

Profiler concurrency

analysis

Parallelism is about performance Step 1: Understand your goals Step 2: Measure existing performance Step 3: Performance tuning

Starts with sequential version Step 4: Identify opportunities for parallelism

Hide latency Speed up CPU bound phases

Step 5: Express parallelism Step 6: Tune

Parallel Application DevelopmentThe basics

Know your users Use realistic datasets Know your ecosystem

Parallelize only if necessary, but PLAN for it Introduces bugs Increases complexity Visual Studio features reduce complexity

I/O, memory latency and algorithmic optimizations Often sufficient to meet goals

Goals And Sequential Tuning

Visual Studio Profiling Tools

demo

Profiler Sampling based – for CPU-bound phases

Lowest overhead Analyze time and/or hardware performance counters

Instrumentation based – not limited to CPU-bound phases Detailed, higher runtime overhead

Managed memory usage analysis

Visual Studio Profiling ToolsDemo recap

TL26: Improving .NET Application Performance And ScalabilitySteve Carroll and Ed Glas

TL24: 1:15 PM in 153 (Wednesday)



Resource manager

Task scheduler


PLINQ


Threads

Operating system

Concurrency runtime

Programming models

Agentslibrary

Thread pool

Task scheduler

Resource manager

Data structures D

ata

stru

ctur

es

Tools

Tools

ParallelDebugger

Toolwindows


analysis

Parallel Application DevelopmentMeasuring performance

Compare actual performance against goals Requires instrumentation

Divide and conquer Identify scenario phases Tackle biggest contributors first Amdahl’s Law

Instrumentation Integrated support in Visual Studio 2010

performance tools

Parallel Application DevelopmentInstrumentation

Managed and native

Instrumentation provides Time ∆, and possibly some resource usage information Nesting support Optional Tracing data emitted for post-run analysis Integration with Visual Studio 2010 Trace Visualization Tools

MeasurementBlock myMB = new MeasurementBlock();myMB.Begin(0, “Phase 1”);

// Phase 1 work happens here

myMB.End();

Parallel Application DevelopmentOpportunities for parallelism

Attempted performance tuning, but can’t meet goals

Identify opportunities for parallelism Speed-up CPU-bound phases Overlap/hide delays from user experience Where and how?

Visual Studio 2010 Concurrency Analysis Tools Managed and native 32-bit and 64-bit Windows 7, Windows Server 2008, and

Windows Vista

Visual Studio 2010 Concurrency Analysis Tools

CPU utilization analysis How it varies with time across phases Identify opportunities for improvement Drill into a range of execution for details

Thread blocking analysis Analyzes each thread’s execution When a thread is not executing, why? Link to source code responsible Disk I/O dependencies with filenames/delays

Core execution analysis Thread migration or affinity issues

Visual Studio Concurrency Analysis Tools

demo

Visual Studio 2010 Concurrency Analysis ToolsWhat I haven't shown

Thread synchronization dependencies Which thread was I waiting for? What was that thread doing?

Multi-process scenarios My program spawns other processes My program requests services from

other processes More statistics and guidance

Point out known issues Provide references to relevant

improvement techniques

Exploiting Parallelism

Forms of parallelism Data, task, pipelining, latency hiding, etc.

Legacy programming patterns Hand-coded thread creation, scheduling Thread pools, I/O completion ports,

asynchronous APIs etc. New programming paradigms

Relieve developer from managing threads Express parallelism using simple abstractions Software stack handles the rest



Resource manager

Task scheduler


Threads

Operating system

Concurrency runtime

Programming models

Agentslibrary

Thread pool

Task scheduler

Resource manager

Data structures D

ata

stru

ctur

es

Tools

Tools

ParallelDebugger

Toolwindows


analysis


PLINQ

Parallelism In .NET Framework 4.0Parallel Extensions

PLINQ – Parallel Language Integrated Query Parallel.For, Parallel.ForEach Parallel.Invoke { stmt1, stmt2,…} Want to learn more?

Parallel Programming For Managed Developers With The Next Version Of Microsoft Visual Studio

Daniel MothTL26: 10:30 AM in Petree Hall CD (Wednesday)


Resource manager

Task scheduler


Threads

Operating system

Concurrency runtime

Programming models

Thread pool

Task scheduler

Resource manager

Data structures D

ata

stru

ctur

es

Tools

Tools

ParallelDebugger

Toolwindows


analysis


PLINQAgentslibrary


Parallel C++ ApplicationsConcurrency runtime and PPL

Parallel Pattern Libraryparallel_for – demo later Parallel task invocations

Asynchronous agentsConcurrency runtime

void quicksort(int * a, int n) { if (n <= 1) return; int s = partition(a,n);

quicksort(a,s); quicksort(a+s,n-s);

return;}

task_group g; g.run([&]{quicksort(a,s);}); g.run([&]{quicksort(a+s,n-s);}); g.wait();

Parallel C++ ApplicationsWant to learn more?

Parallel Programming For C++ Developers In The Next Version Of Microsoft Visual Studio

Rick Molloy TL25: 3:30 PM in 408B (TODAY)

Concurrency Runtime Deep Dive: How To Harvest Multicore Computing Resources

Niklas GustafssonTL22: 1:15 PM in 408B (Wednesday)

Expressing ParallelismRemaining developer tasks

Ensure that parallel tasks can execute concurrently Without violating order or data dependencies Example: parallel_for loop iterations

have to be independent Synchronization

Identify races and ordering constraints Add appropriate synchronization primitives

Visual Studio Concurrent Programming And Tool Support

demo

Call To ActionConcurrency in Visual Studio 2010

Many new features for parallel programmers Explore these in the Visual Studio CTP Give us your feedback!

Programming constructs to reduce program complexity Managed and native Concurrency runtime Task Parallel library Parallel Pattern library

Concurrency analysis tools Analyze concurrency in your application Understand sources of inefficiency Support for TPL and PPL

Questions? Concurrency Runtime: TL22: 1:15 PM in 408B (Wednesday).NET Performance: TL24: 1:15 PM in 153 (Wednesday)Native Parallel Programming: TL25: 3:30 PM in 408B (Today)Managed And Debugging: TL26: 10:30 AM in Petree Hall (Wednesday)Parallel Computing Symposium: 8:30 AM – 1:30 PM 515A (Thursday)

http://blogs.microsoft.com/hshafi

http://blogs.microsoft.com/hshafi

Evals & Recordings

Please fill

out your

evaluation for

this session at:

This session will be available as a recording at:

www.microsoftpdc.com

© 2008 Microsoft Corporation. All rights reserved. Microsoft, Windows, Windows Vista and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market

conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.

Hazim Shafi Principal Architect Microsoft Corporation TL19.

Documents

Transcript of Hazim Shafi Principal Architect Microsoft Corporation TL19.