.NET Parallel Programming in C#

1. .NET Parallel Programming in C#

2. Three approaches • There are three approaches you can run your tasks in parallel • Task programming • For control parallel • Parallel Loops • For data parallel • Parallel LINQ • For concurrency based on concurrent data access

3. Task programming—start tasks task1.Start(); //Thread.Sleep(1000); task2.Start(); //Thread.Sleep(1000); task3.Start(); //Thread.Sleep(1000); task4.Start(); // wait for input before exiting Console.WriteLine("Main method complete. Press enter to finish."); Console.ReadLine(); } static void printMessage(inti) { //Thread.Sleep(1000); Console.WriteLine("Hello World, this is task" + i.ToString() + " with id = " + Task.CurrentId.ToString() + " at " + System.DateTime.Now.ToString()); } static void printMessage() { printMessage(1); } } } using System; using System.Threading.Tasks; using System.Threading; namespace Listing_02 { class Listing_02 { static void Main(string[] args) { // use an Action delegate and a named method Task task1 = new Task(new Action(printMessage)); // use a anonymous delegate Task task2 = new Task(delegate { printMessage(2); Console.WriteLine("This is task2"); }); // use a lambda expression and a named method Task task3 = new Task(() => printMessage(3)); // use a lambda expression and an anonymous method Task task4 = new Task(() => { printMessage(4); Console.WriteLine("This is task4"); });

4. Task programming— passing an object task1.Start(); task2.Start(); task3.Start(); task4.Start(); // wait for input before exiting Console.WriteLine("Main method complete. Press enter to finish."); Console.ReadLine(); } static void printMessage(object message) { Console.WriteLine("Message: {0}", message + " from task with id " + Task.CurrentId.ToString()); } } } using System; using System.Threading.Tasks; namespace Listing_03 { class Listing_03 { static void Main(string[] args) { // use an Action delegate and a named method Task task1 = new Task(new Action<object>(printMessage), "First task"); // use a anonymous delegate Task task2 = new Task(delegate(object obj) { printMessage(obj); }, "Second Task"); // use a lambda expression and a named method Task task3 = new Task((obj) => printMessage(obj), "Third task"); // use a lambda expression and an anonymous method Task task4 = new Task((obj) => { printMessage(obj); }, "Fourth task");

5. Task programming—passing an object (2) string[] messages = { "Task 1", "Task 2", "Task 3", "Task 4", "Task 5", "Task 6", "Task 7", "Task 8", "Task 9", "Task 10" }; foreach(string msg in messages) { Task myTask = new Task(obj => printMessage((string)obj), msg); myTask.Start(); }

6. Task programming—getting results // create the task using state Task<int> task2 = new Task<int>(obj => { int sum = 0; int max = (int)obj; for (int i = 0; i < max; i++) { sum += i; } return sum; }, 100); // start the task task2.Start(); // write out the result Console.WriteLine("Result 2: {0}", task2.Result); // wait for input before exiting Console.WriteLine("Main method complete. Press enter to finish."); Console.ReadLine(); } } } using System; using System.Threading.Tasks; namespace Listing_05 { class Listing_05 { static void Main(string[] args) { // create the task Task<int> task1 = new Task<int>(() => { int sum = 0; for (int i = 0; i < 100; i++) { sum += i; } return sum; }); // start the task task1.Start(); // write out the result Console.WriteLine("Result 1: {0}", task1.Result);

7. Task programming—getting results • It is like the return value of a method • Imaging using this approach to get the max of an array segment • How do we know all the tasks are done?

8. Task programming—waiting • Wait for one • Task.Wait(); // wait for completes, is canclled, or throws an exception • Task.Wait(CancellationToken); //cancelled or completes • Task.Wait(Int32); // wait for Int32 milliseconds or Wait() • Task.Wait(TimeSpan); //TimeSpan represent length of time • Task.Wait(Int32, CancellationToken) • Wait for any • wait for all • See list 6

9. Task programming—other activities • Cancel a task • Monitoring calculation • Waiting for time to pass • Handling exceptions • Getting task status • Task continuations • A way to coordinate tasks • Conclusions • Task programming can achieving concurrency • It supports control parallelism better • When you only have a few cores, this is a good tool

10. One more • System.Collections.Concurrent

11. Parallel Loops • There are two • Parallel.For, and • Parallel.ForEach (see an example 5-4) // create a collection of strings List<string> dataList = new List<string> { "the", "quick", "brown", "fox", "jumps", "etc" }; // process elements of the collection using a parallel foreach loop Parallel.ForEach(dataList, item => { Console.WriteLine("Item {0} has {1} characters", item, item.Length); });

12. Parallel Loop Options

13. ParallelLoopState Class • Enables iterations of Parallel loops to interact with other iterations. An instance of this class is provided by the Parallel class to each loop; you can not create instances in your user code.

14. ParallelLoopStateClass (2) • Two Key methods • You can get the result if you break/stop (5-9)

15. Using Thread Local Storage • Interlocked is a global level lock, the more cores you have the more expensive it becomes • Still, it is better than the lock object • Thread Local Storage is a better solution • Example 5 - 11 Parallel.For( 0, 100, () => 0, (int index, ParallelLoopStateloopState, inttlsValue) => { tlsValue += index; return tlsValue; }, value => Interlocked.Add(ref total, value));

16. Partitioning • This one answers how to divide works and associated data among tasks • The trick is to reduce synchronization overhead while keeping the workload balanced • Generally speak, there are two approaches • Static • Dynamic • PLINQ uses static • We will look into dynamic here

17. Partitioning (2) • Must use ForEach • You can create your own partitioning schema