Coding For Cores - C# Way

Coding For Cores - C# Way

• 1.
  Introduction Multicore processors havebeen around for many years. However, many developers are doing what they’ve always done: creating single-threaded programs. They’re not taking advantage of all the extra processing power. Example: Imagine you have many tasks to perform and many people to perform them, but you are using only one person because you don’t know how to ask for more. It’s inefficient. Users are paying for extra power, but their software is not allowing them to use it.
• 2.
  Free lunch isover • Don’t expect your sequential program to run faster on new processors. • Today’s desktops typically have 4 cores. Latest Intel multi-core chip has 48 cores. Expect 100s of cores in the near future. • To take advantage of the hardware of today and tomorrow, we have to parallelize our code to distribute work across multiple processor cores. • The .NET Framework provides several ways to use multiple threads of execution to keep application responsive to our user while maximizing the performance.
• 3.
  Parallel Programming inthe .NET Framework • In the past, parallelization required low-level manipulation of threads, thread pools and locks. • Visual Studio 2010 and the .NET Framework 4.0: • new runtime • new class library types • new diagnostic tools • Features simplify parallel development to write efficient, fine-grained, and scalable parallel code.
• 4.
  Parallel programming architecture.NET4: High-level overview
• 5.
  Task Parallel Library(TPL) • Is a set of public types and APIs in the System.Threading and System.Threading.Tasks namespaces. • The TPL scales the degree of concurrency dynamically to most efficiently use all the processors that are available. • TPL handles the partitioning of the work, the scheduling of threads on the threadpool, cancellation support, state management, and other low-level details. • To remember: • Not all code is suitable for parallelization; e.g. for small/cheap loops, overhead of parallelization can cause the code to run more slowly. • Parallelization adds complexity to program execution. • Need understanding of threading concepts: locks, deadlocks, and race conditions, to use TPL effectively.
• 6.
  Data Parallelism (TPL) •Data parallelism refers to scenarios in which the same operation is performed concurrently (that is, in parallel) on elements in a source collection or array. • Through System.Threading.Tasks.Parallel class. • Imperative way: • Describes how to create parallel for and foreach loops. • When to use: Strategy can work well if you have either of these • Lots of items • Lots of work for each item • DEMO
• 7.
  Data Parallelism... • Declarativeway: PLINQ (Parallel LINQ) • PLINQ)is a parallel implementation of the LINQ pattern • Partition the data source into segments, and then executing the query on each segment on separate worker threads in parallel on multiple cores. • PLINQ can achieve significant performance improvements over legacy code for many cases. • However, parallelization complexity can actually slows down certain queries, if not used wisely.
• 8.
  Task Parallelism (TPL) •Based on the concept of a task, which represents an asynchronous operation. • A task resembles a thread or ThreadPool work item, but at a higher level of abstraction. • The term task parallelism refers to one or more independent tasks running concurrently.
• 9.
  Task benefits… • Moreprogrammatic control than is possible with a thread: waiting, cancellation, continuations, robust exception handling, detailed status, custom scheduling, and more. • More efficient and scalable use of system resources: tasks are queued to the ThreadPool, enhanced with algorithms that determine and adjust to the number of threads and that provide load balancing to maximize throughput. This makes tasks relatively lightweight. • For above reasons: TPL is the preferred API for writing multi- threaded, asynchronous, and parallel code.
• 10.
  Creating and RunningTasks • Implicit: • Parallel.Invoke method to run any number tasks concurrently. • Action delegate is passed for each item of work (task). • Syntax: Parallel.Invoke(() => DoSomeWork(), () => DoSomeOtherWork()); • Demo ( parallelizes the operations, not the data) • Explicit: • Create Task or Task<TResult> with user delegate (named delegate, anonymous method, or lambda expression) that encapsulates the code task will execute. • Execute with task1.Start() with other functions (task1.Wait() etc) • Task.Run() methods to create and start a task in one operation • Preferred way when more control over the creation and scheduling of the task is not needed.
• 11.
  Tasks... • The tasksrun asynchronously and may complete in any order. If the Result property is accessed before the computation finishes, the property blocks the calling thread until the value is available.
• 12.
  Options • C# providesseveral mechanisms for parallel programming: • Explicit threads: • with synchronization via locks, critical regions etc. • The user gets full control over the parallel code. • BUT orchestrating the parallel threads is tricky and error prone (race conditions, deadlocks etc) • This technique requires a shared-memory model • Explicit threads with a message-passing library: • Threads communicate by explicitly sending messages, with data required/produced, between workstations. Parallel code can run on a distributed-memory architecture, eg. a network of workstations. The programmer has to write code for (un-)serializing the data that is sent between machines. BUT threads are still explicit, and the difficulties in orchestrating the threads are the same. A common configuration is C+MPI.
• 13.
  Wait a minute...Quantum Computation on a way! • Calls modern day computing classical one. • Uses quantum superposition principle on superconductors. • Can do: • Integer prime factorization (security of public key cryptographic systems) • Quadratic speedup over NP-complete problems (Viz. TSP, Subset sum problem etc.) • Much more... • Example: D-wave 2x system
• 14.
  Resources • https://github.com/bsonnino/ParallelProgramming • QC:Dwave, ScienceDaily, Wikipedia