The server side story: Parallel and Asynchronous programming in .NET - ITProDevConnections 2014
Download as PPTX, PDF0 likes75 views
This document discusses parallel and asynchronous programming on servers. It covers techniques like Task Parallel Library (TPL), Reactive Extensions (Rx), and Dataflow that can be used for parallel processing on servers. Unlike desktop applications where the focus is on reducing execution time, server applications prioritize throughput and scalability over individual request duration. Asynchronous programming is more important on servers to avoid blocking and improve throughput. The document demonstrates various asynchronous programming patterns on ASP.NET like async actions and background processing using libraries like SignalR. It also provides demos of parallel programming techniques like Parallel.For, TPL Dataflow, and Rx.
The server side story: Parallel and Asynchronous programming in .NET - ITProDevConnections 2014
- 1. Parallel and Asynchronous Programming The server-side story Panagiotis Kanavos
- 2. Parallel Extensions • Best kept .NET secret known by all • Since 2010 and .NET 4.0 • Coves Task, Data based processing models • Close cousins • TPL Dataflow • System.Reactive • Available through NuGet • SIMD through RyuJIT
- 3. Use on the desktop • async/await for asynchronous events • Task.Run for background Processing • Parallel computation • Priorities • Avoid blocking the UI • Reduce execution time • Use all available cores
- 4. What’s different on the Server • No UI thread • Multiple concurrent users/requests • Throughput/Scalability more important than request durattion • Then there are timeouts • Lots of services, lots of combined timeouts • And …
- 6. Server Meltdown • Load Balanced farm • High load • First server stuck at 100% CPU • Timeouts, App pool Recycle • More requests redirected to other servers • Second server reaches 100% • Don’t drop the ball! • …
- 7. Request Queuing in ASP.NET IIS Queue • Req 1 • Req 2 • Req 3 ASP.NET Worker Threads • Thread 1 • Thread 3 App Domain • Thread 2
- 8. Why … • Lots of threads Worse Scalability • Can’t run more threads than cores • Thrashing • When everyone awaits everyone else, no-one finishes • HATES blocking calls! • Thread Pool starvation Can’t serve • 500 Too Busy • Immediate actio • App Pool Recycle • Only possible action really …
- 9. Asynchronous Execution • More important for server than speed • Stackless Python No blocking calls • A view of the real world • No blocking I/O at the OS level • I/O = File, Database, WS/REST calls • On IIS – IO Threads • Since the IO Thread is doing the job, why am I waiting for it?
- 10. Async on WebForms • Basic support through async event handlers • async void Button_Click μπρρρ… • <%@ Page Async=“true” %> • Use Page.RegisterAsyncTask for jobs longer than a single request • Raw Background threads can be killed at any time
- 11. Async on MVC/Web API • Async actions! • Plays well with Database/EF async • HostingEnvironment.QueueBackgroundWorkItem • Notifies IIS • Gives 90 sec to task to finish in case of shutdown/recycle • HangFire, Azure Web Jobs for repeating tasks
- 12. Parallel.For and PLINQ • Data Parallelism • Cuts the data in chunks • Task per chunk • Generally, 1 Task/Core • Also uses the calling thread • Looks like blocking (it’s not) • Not a concern on the server
- 13. Dataflow DB call • I:Date • O:Sales Web call • I:Sale • O:Details • X3 • Bounded Analyze Store • Bulk Insert to DB
- 14. Dataflow – Old Friend
- 15. The basics • Processing models • Agent and Pipeline • and MapReduce data processing • One task per block (configurable) • Input/Output Buffer • Block Types • Buffering • Execution • Grouping • Links between blocks
- 16. Execution Blocks • Action Block • Classic Job Queue • Rudimentary Agent implementation • Transform Block • Transform one Input item to one Output • Call a web service for a Stock to get History as a package • TransformMany Block • For one input item multiple output items • Parallel Execution
- 17. Throttling • BoundedCapacity • Maximum input size • Blocks sources • MaxMessagesPerTask • Recycle a Task after X messages • CPU Throttling
- 19. Buffering Blocks • Buffer Block • Easy Publisher/Subscriber implementation • Broadcast Block • Last event sent to many subscribers • WriteOnce block
- 20. Batch Block Demo
- 21. Grouping Blocks • Batching Block • Join Block • 2-3 Inputs, 1 Tuple<T1,T2,T3> output • BatchedJoin Block • Batch Join • T1, T2 Tuple< IList<T1>, IList<T2>>
- 23. Links • LinkTo IDisposable • Disconnect Dispose • Settings • Filter messages through Lambda • Propagate Completion
- 24. Custom combination with Encapsulate • New block • One Input, One Output
- 25. Error Handling In case of error a block • Discards input messages • Refuses new messages • Propagate exception • Caught with await/.Wait()
- 26. Dataflow Demos
- 27. Reactive Extensions • Event monitoring and processing • Control Events • Records in Event Log • Calls to a Web Service • Time-based events • Handle as a stream • Works like LINQ • Foundation for Event Processing
- 28. Sources • Anything implementing IObservable<> • Time • Observable.Interval • Events • Observable.FromEventPattern • BeginXXX/EndXXX • Observable.FromAsyncPattern • State machines / “Loops” • Observable.Generate • Tasks • Every IEnumerable<T> • Subject<T> mySubject.OnNext(data)
- 29. Reactive Demos
- 30. Useful Functions • Interval • Buffer • Sample • Throttle • Window • Skip • Take • TakeUntil
- 31. Transformations • Anything goes • task ... AsObservable() • block … AsObservable() • await Observable • ToTask<T>
- 32. SignalR • Real Time communication from Server to Client • Push Notifications • Long process progress • Eg. Search for tickets
- 33. Multiple Techniques • Web sockets, falling back to … • Server Sent Events • Forever Frame • Ajax Long Polling
- 34. Call function on Browser • Server-side: Clients.All.DoSomething(data) • Browser-side hub.client.updateProgress = function (data)
- 38. Supported backplanes • Windows Azure Bus • SQL Server • Redis • Recommended Distributed memory cache for new Azure projects • Open Source Rulez!
- 39. SignalR Demos
- 40. Useful links- Courses MVA Course • Lighting up Real-Time Web Communications with SignalR Pluralshight Course • Async and Parallel Programming: Application Design Books • The C# Concurrency Cookbook, Stephen Cleary, O’Reilly
- 41. Useful Links - Sites Sites • Signal R • TPL Dataflow • Reactive Extensions • Using Asynchronous Methods in ASP.NET • 101 Rx Samples Blogs • Stephen Cleary • The Magic of using Acync in ASP.NET, Scott Hanselman • How to run background tasks in ASP.NET, Scott Hanselman
- 42. Session Evaluation Your feedback is important and valuable Submit before the event’s close session to WIN prizes 2ways to access Go to m.itprodevconnections.gr Ask for an Evaluation Sheet from the registration desk
- 43. Microsoft Virtual Academy for IT Pros Όλα τα trainings που χρειάζεσαι δωρεάν σε ένα site! /virtualization/ /server-infrastructure/ /private-cloud/ /hybrid-cloud/ /desktop-devices-management/ /licensing/ /και πολλά άλλα… www.microsoftvirtualacademy.com MVA
- 44. Microsoft Virtual Academy LIVE Όλα τα trainings που χρειάζεσαι δωρεάν σε ένα site! www.microsoftvirtualacademy.com 4 ΜΕΡΕΣ Εκπαίδευση με το Mark Russinovich ΔΩΡΕΑΝ ΠΡΟΕΤΟΙΜΑΣΙΑ για την Εξέταση 70-533: Implementing Azure Infrastructure Solutions Και 50% ΕΚΠΤΩΤΙΚΟ VOUCHER ΣΤΟ ΚΟΣΤΟΣ ΤΗΣ ΕΞΕΤΑΣΗΣ 1 ως 4 Δεκεμβρίου MVA
- 45. Microsoft Virtual Academy for Devs Όλα τα trainings που χρειάζεσαι δωρεάν σε ένα site! /game development / /web development / /cloud development / /mobile development / /C#-XAML-HTML/ /app development/ /visual studio και πολλά άλλα… www.microsoftvirtualacademy.com MVA