Parallel programming in .NET
- 1. Peter Csala
- 2. 2 Agenda • Definitions & Terminologies • Problems & Toolbox • Demos
- 4. 4 Concepts & Misconceptions Parallel Programming Concurrent Programming Asyncronous Programming Multi-Threaded Programming
- 5. 5 Prog. model: Sync vs Async
- 6. 6 Execution: Parallel vs Concurrent
- 7. 7 Environment: Single vs Multi Threading
- 8. 8 Rumor: Zen 3 can run 4 threads / core
- 9. 9 Rumor: Ryzen ThreadRipper (64cores)
- 10. 10 Types: Data vs Task
- 11. Problems & Tools
- 13. 13 Toolbox against concurrent mutation • Syncronization Primitives • Concurrent Collections • Immutable state
- 14. 14 Patterns against concurrent mutation • Fork / Join • Master – Worker • Producer + Consumer • Divide & Conquer • MapReduce • Observer • Pipeline • All To All • Stencil • Etc.
- 15. 15 It’s all about isolation
- 16. Enough theory, Let’s practice
- 17. 17 The Downloader Functional Req.s • I want to be able to download arbitrary number of web contents • I want to be able to process them in arbitrary way • It MUST be blazingly fast • It MUST be superb easy to use • Why is this not done yet? Non-Functional Req.s • Ensure that the provided strings are valid Uris • Ensure post-processing ability via continuation / callback • It should run in parallel without exhausting all resources • Interface should ask for only the bear minimum (use default params)
- 23. 23 Architecture: Shared State • The whole source is shared • The source is pre-partitioned • The partitions are processed separately by pulling
- 24. 24 Architecture: Shared State • The source is pre-partitioned • The partitions are not shared • The partitions are processed separately by iterating
- 25. 25 Architecture: Shared State • The whole source is shared • The source is dynamically partitioned • The source pushes the data to the targets
- 27. 27 What have we learnt?
- 28. 28 What have we learnt? • (Non-)deterministic • Slow down • Over parallelisation • Partitioning is hard • Common pitfalls • Measure base line • Measure speed-up • CPU vs I/O bound • Immutables • Enormous toolset
- 29. 29 Where should I look for details? • Parallel class • PLINQ partitioning • Channels • Parallel data structure, 2 • Synchronization primitives, 2 • Compare Parallel models • Immutable Collections, 2 • Throttling • Potential Pitfalls, PLINQ • Async Guidance • Async antipatterns • Avoid sync problem • VS Threading nuget • Concurrency Visualizer
- 30. 30 Is there any free e-book? • Patterns & Practices Parallel Programming with Microsoft .NET • Patterns of Parallel Programming • Introduction of TPL Dataflow • Parallel Extension Extras
- 31. 31 Is there any good book?
Editor's Notes
- #5: https://medium.com/swift-india/concurrency-parallelism-threads-processes-async-and-sync-related-39fd951bc61d
- #6: http://www.enterpriseintegrationpatterns.com/patterns/messaging/Introduction.html
- #7: https://www.raywenderlich.com/60749/grand-central-dispatch-in-depth-part-1
- #8: https://www.cs.uic.edu/~jbell/CourseNotes/OperatingSystems/4_Threads.html
- #9: https://www.tweaktown.com/news/67801/amds-next-gen-zen-3-rumor-4-threads-per-core-enables-64c-256t-chips/index.html
- #10: https://wccftech.com/amd-ryzen-threadripper-3960x-24-core-48-thread-7nm-zen-2-cpu-confirmed/
- #11: http://help.pervasive.com/display/DR62/General+Concepts
- #15: https://onlinelibrary.wiley.com/doi/full/10.1002/cpe.4175 http://snir.cs.illinois.edu/patterns/patterns.pdf