New Microsoft case studies on Parallel Programming

Below is some links of Microsoft case studies of Parallel Programming implementation.

Composite

http://www.microsoft.com/casestudies/Case_Study_Detail.aspx?casestudyid=4000006833

ORELogy

http://www.microsoft.com/casestudies/Case_Study_Detail.aspx?casestudyid=4000006754

Visual Numerics

http://www.microsoft.com/casestudies/Case_Study_Detail.aspx?casestudyid=4000006177

Grange Insurance

http://www.microsoft.com/casestudies/Case_Study_Detail.aspx?casestudyid=4000006894

Task Parallelism: Parallel programming – II

 

Please read Begin with Parallel programming in Dotnet 4.0 article if you have not read.

Task Parallelism

This is strategy to run independent task in parallel way. It focuses on distributing execution process (threads) across different parallel computing nodes.Task parallelism emphasizes the distributed (parallelized) nature of the processing (i.e. threads), as opposed to the data parallelism. Most real programs fall somewhere on a continuum between Task parallelism and Data parallelism. Workflow of task parallelism is below:

 

image

Dot Net Framework provides Task Parallel Library (TPL)  to achieve Task Parallelism. This library provides two primary benefits:

  1. More Efficient and more scalable use of system resources.
  2. More programmatic control than is possible with a thread or work item.

Behind the scenes tasks are queued in ThreadPool, which has been enhanced with algorithms in .Net 4.0 that determine and adjust to the number of threads that maximizes throughput. This makes tasks relatively lightweight, and you can create many of them to enable fine-grained parallelism. To complement this, widely-known work-stealing algorithms are employed to provide load-balancing.

This library provides more features to control tasks like: cancellation, continuation, waiting, robust exception handling, scheduling etc.

The classes for TaskParallelism are defined in System.Threading.Tasks:

Class Purpose
Task For running unit of work concurrently.
Task<Result> For managing unit of work with return value
TaskFactory Factory Class to create Task class Instance.
TaskFactory<TResult> Factory Class to create Task class Instance and return value.
TaskScheduler for scheduling tasks.
TaskCompletionSource For manually controlling a
task’s workflow

How to Create and execute Tasks

Task Creation and Execution can be done by two ways: Implicit and Explicit.

Create and Execute Task Implicitly

Parallel.Invoke method helps to to run unit of work in parallel. you can just pass any number of Action delegates as parameters. The no. of tasks created by Invoke method is not necessarily equal to Action delegates provided because this method automatically does some optimization specially in this case.

Source Code

        private static void Run2()
        {
            Thread.Sleep(1000);
            Console.WriteLine("Run2: My Thread Id {0}", Thread.CurrentThread.ManagedThreadId);
        }

        private static void Run1()
        {
            Thread.Sleep(1000);
            Console.WriteLine("Run1: My Thread Id {0}", Thread.CurrentThread.ManagedThreadId);
        }

        static void Main(string[] args)
        {
            //Create and Run task implicitly
            Parallel.Invoke(() => Run1(), () => Run2());
            Console.ReadLine();
        }

Output

Run2: My Thread Id 11

Run1: My Thread Id 10

This approach of creating task does not give greater control over task execution, scheduling etc. Better approach is to create task by TaskFactory class.

Create and Execute Task Explicitly

You can create task by creating instance of task class and pass delegate which encapsulate the code that task will execute. These delegate can be anonyms, Action delegate, lambda express and method name etc.

Example of creating tasks:

Source Code:

        private static void Run2()
        {
            Thread.Sleep(1000);
            Console.WriteLine("Run2: My Thread Id {0}", Thread.CurrentThread.ManagedThreadId);

        }

        private static void Run1()
        {
            Thread.Sleep(1000);
            Console.WriteLine("Run1: My Thread Id {0}", Thread.CurrentThread.ManagedThreadId);

        }

        static void Main(string[] args)
        {
            // Create a task and supply a user delegate by using a lambda expression.
            // use an Action delegate and a named method
            Task task1 = new Task(new Action(Run1));

            // use a anonymous delegate
            Task task2 = new Task(delegate
                        {
                            Run1();
                        });

            // use a lambda expression and a named method
            Task task3 = new Task(() => Run1());
            // use a lambda expression and an anonymous method
            Task task4 = new Task(() =>
            {

                Run1();
            });

            task1.Start();
            task2.Start();
            task3.Start();
            task4.Start();
            Console.ReadLine();
        }

Output

Run1: My Thread Id 13

Run1: My Thread Id 12

Run1: My Thread Id 11

Run1: My Thread Id 14

If you don’t want to create and starting of task separated then you can use TaskFactory class. Task exposes “Factory” property which is instance of TaskFactory class.

Task task5= Task.Factory.StartNew(()=> {Run1();});

Task with Return Values

To get value from when task completes it execution, you can use generic version of Task class.

 public static void Main(string[] args)
        {
            //This will return string result
            Task task = Task.Factory.StartNew(() => ReturnString());
            Console.WriteLine(task.Result);// Wait for task to finish and fetch result.
            Console.ReadLine();
            
        }

        private static string ReturnString()
        {
            return "Neeraj";
        }

Task State

If you are running multiple tasks same time and you want to track progress of each task then using "State" object is better approach.

 
	public static void Main(string[] args)
        {
            //This will return string result
            for (int i = 0; i < 5; i++)
            {
                Task task = Task.Factory.StartNew(state => ReturnString(), i.ToString());
                //Show progress of task
                Console.WriteLine("Progress of this task {0}: {1}", i, task.AsyncState.ToString());
            }
            
            Console.ReadLine();

        }

        private static void  ReturnString()
        {
            //DO something here

           // Console.WriteLine("hello");
        }

Output

Progress of this task 0: 0

Progress of this task 1: 1

Progress of this task 2: 2

Progress of this task 3: 3

Progress of this task 4: 4

In Next blog, I’ll explain TaskCreationOptions,Waiting,Cancellation tasks etc.

Please keep giving your valuable feedbacks.

Begin with Parallel programming in Dotnet 4.0

 

Now a days computers are coming with multiple processors that enable multiple threads to be executed simultaneously to give performance of applications and we can expect significantly more CPUs in near future. If application is doing CPU intensive tasks and we find that one CPU is taking 100 %usage and others are idle. It might be situation when one thread is doing cpu intensive work and other threads are doing non cpu intensive work. In this case application is not utilizing all CPUs potential here. To get benefits all CPUs Microsoft launches Parallel Programming Library in DotNet Framework 4.0.

We can say “Programming to leverage multicores or multiple processors is called parallel programming”. This is a subset of the broader concept of multithreading.

To use your system’s CPU resources efficiently, you need to split your application into pieces that can run at the same time and we have CPU intensive task that should be breaks into parts like below:

  1. Partition it into small chunks.
  2. Execute those chunks in parallel via multithreading.
  3. Collate the results as they become available, in a thread-safe and performant manner.

We can also achieve this in traditional way of multithreading but partitioning and collating of chunks can be nightmare because we would need to put locking for thread safety and lot of synchronizatopm to collate everything. Parallel programming library has been designed to help in such scenarios. You don’t need to worry about partitioning and collating of chunks. These chunks will run in parallel on different CPUs.

There can be two kind of strategy for partitioning work among threads:

  1. Data Parallelism: This strategy fits in scenario in which same operation is performed concurrently on elements like collections, arrays etc.
  2. Task Parallelism: This strategy suggests independent tasks running concurrently.

Data Parallelism

In this parallelism, collection or array is partitioned so that multiple threads can operate on different segments concurrently. DotNet supports data parallelism by introducing System.Threading.Tasks.Parallel static class. This class is having methods like Parallel.For, Parallel.ForEach etc. These methods automatically partitioned data on different threads, you don’t need to write explicit implementation of thread execution.

Below is simple example of Parallel.For Method and you can see how it has utilize all cores and you are not using any synchronization here:

Parallel.For(0, 20, t => { Console.WriteLine("Thread{0}:Value:{1}",Thread.CurrentThread.ManagedThreadId,t); });

Output

Thread10:Value:0
Thread7:Value:5
Thread10:Value:1
Thread6:Value:10
Thread10:Value:2
Thread10:Value:3
Thread10:Value:4
Thread10:Value:8
Thread10:Value:9
Thread10:Value:13
Thread10:Value:14
Thread10:Value:16
Thread10:Value:17
Thread10:Value:18
Thread10:Value:19
Thread12:Value:15
Thread6:Value:11
Thread6:Value:12
Thread7:Value:6
Thread7:Value:7

Above example tells you how Parallel class is utilizing all cores. Now I am giving you example of performance of Parallel loop for CPU intensive tasks over sequential loop.

System.Diagnostics.Stopwatch stopwatch = new System.Diagnostics.Stopwatch();
    //Sequential Loop
    stopwatch.Start();
    for(int i=0;i<20000;i++)
        {
            double temp = i * new Random().Next() + Math.PI;
        }
    stopwatch.Stop();
    Console.WriteLine("Total Time (in milliseconds) Taken by Sequential loop: {0}",stopwatch.ElapsedMilliseconds);
    stopwatch.Reset();

    //Parallel Loop
    stopwatch.Start();
    Parallel.For(0, 20000, t => 
        {
            double temp = t * new Random().Next() +Math.PI;
        });
    stopwatch.Stop();
    Console.WriteLine("Total Time (in milliseconds) Taken by Parallel loop: {0}", stopwatch.ElapsedMilliseconds); 

Output

 

Total Time (in milliseconds) Taken by Sequential loop: 159
Total Time (in milliseconds) Taken by Parallel loop: 80

 


I’ll explain Task Parallelism in next blog. Please share your comments.