[C#] 並行編程Parallel

一、Parallel的使用

在Parallel下麵有三個常用的方法Invoke、For和ForEach。

1、Parallel.Invoke

 1 public class ParallelDemo
 2 {
 3     private Stopwatch stopWatch = new Stopwatch();
 4 
 5     public void Run1()
 6     {
 7         Thread.Sleep(2000);
 8         Console.WriteLine("Task 1 is cost 2 sec");
 9     }
10     public void Run2()
11     {
12         Thread.Sleep(3000);
13         Console.WriteLine("Task 2 is cost 3 sec");
14     }
15 
16     public void ParallelInvokeMethod()
17     {
18         stopWatch.Start();
19         Parallel.Invoke(Run1, Run2);
20         stopWatch.Stop();
21         Console.WriteLine("Parallel run " + stopWatch.ElapsedMilliseconds + " ms.");
22 
23         stopWatch.Restart();
24         Run1();
25         Run2();
26         stopWatch.Stop();
27         Console.WriteLine("Normal run " + stopWatch.ElapsedMilliseconds + " ms.");
28     }
29 }                

執行Parallel.ParallelInvokeMethod()的結果：

可以看出正常調用Run1()和Run2()所花費的時間在5秒左右，而用Parallel.Invoke方法調用值用了3秒左右，也就是耗時最長的那個方法所花費的時間。

2、Parallel.For

這個方法和For迴圈類似。在ParallelDemo類中添加一個ParallelForMethod()方法：

 1 public void ParallelForMethod()
 2         {
 3             stopWatch.Start();
 4             for (int i = 0; i < 10000; i++)
 5             {
 6                 for (int j = 0; j < 60000; j++)
 7                 {
 8                 }
 9             }
10             stopWatch.Stop();
11             Console.WriteLine("NormalFor run " + stopWatch.ElapsedMilliseconds + " ms.");
12 
13             stopWatch.Reset();
14             stopWatch.Start();
15             Parallel.For(0, 10000, item =>
16             {
17                 for (int j = 0; j < 60000; j++)
18                 {
19                 }
20             });
21             stopWatch.Stop();
22             Console.WriteLine("ParallelFor run " + stopWatch.ElapsedMilliseconds + " ms.");
23 
24         }

運行結果：

這個例子中Parallel.For比for耗費的時間要少，但是Parallel.For並不總是比for快。例如：

 1             long sum = 0;
 2             stopWatch.Start();
 3             for (int i = 0; i < 10000; i++)
 4             {
 5                 for (int j = 0; j < 60000; j++)
 6                 {
 7                     sum++;
 8                 }
 9             }
10             stopWatch.Stop();
11             Console.WriteLine("NormalFor run " + stopWatch.ElapsedMilliseconds + " ms.");
12 
13             stopWatch.Reset();
14             stopWatch.Start();
15             Parallel.For(0, 10000, item =>
16             {
17                 for (int j = 0; j < 60000; j++)
18                 {
19                     sum++;
20                 }
21             });
22             stopWatch.Stop();
23             Console.WriteLine("ParallelFor run " + stopWatch.ElapsedMilliseconds + " ms.");

 用Parallel.For所耗費的時間反而更多，是for的3倍以上，這是因為Parallel.For是並行運行的，所以會同時訪問sum，會出現資源爭奪，耗費大量的時間在資源等待上面。

3、Parallel.ForEach

Parallel.ForEach與foreach類似

 1         public void ParallelForEachMethod()
 2         {
 3             int[] nums = new int[100];
 4             for (int i = 0; i < 100; i++)
 5             {
 6                 nums[i] = i;
 7             }
 8             stopWatch.Start();
 9             foreach (var item in nums)
10             {
11                 Console.Write(item + "\t");
12             }
13             stopWatch.Stop();
14             Console.WriteLine("\nNormalFor run " + stopWatch.ElapsedMilliseconds + " ms.");
15 
16             stopWatch.Reset();
17             stopWatch.Start();
18             Parallel.ForEach(nums, item =>
19             {
20                 Console.Write(item + "\t");
21             });
22             stopWatch.Stop();
23             Console.WriteLine("\nParallelFor run " + stopWatch.ElapsedMilliseconds + " ms.");
24         }

從0到99輸出，從運行結果可以看出，Parallel.ForEach的輸出順序是不定的，而foreach是按順序輸出。

二、Parallel中途退出和異常處理

1、在串列操作中，使用break語句即可退出迴圈，但是再並行操作中，就不是這麼回事了。並行迴圈的委托參數中含有一個ParallelLoopState，該實例擁有Break()和Stop()這兩個方法，可以用來退出並行迴圈。

Break：當然這個是通知並行計算儘快的退出迴圈，比如並行計算正在迭代100，那麼break後程式還會迭代所有小於100的。

Stop：這個就不一樣了，比如正在迭代100突然遇到stop，那它啥也不管了，直接退出。

下麵一段代碼測試：

 1         public void ParallelBreak()
 2         {
 3             ConcurrentBag<int> bag = new ConcurrentBag<int>();
 4             Stopwatch stopWatch = new Stopwatch();
 5             stopWatch.Start();
 6             Parallel.For(0, 1000, new ParallelOptions { MaxDegreeOfParallelism = 1 }, (i, state) =>
 7             {
 8                 if (bag.Count == 300)
 9                 {
10                     state.Stop();
11                     return;
12                 }
13                 bag.Add(i);
14             });
15             stopWatch.Stop();
16             Console.WriteLine("Bag count is " + bag.Count + ", " + stopWatch.ElapsedMilliseconds);
17         }

當Parallel.For迴圈遍歷了300個元素時，就運行Stop()方法退出迴圈。

2、異常處理

 1         public void Run1()
 2         {
 3             Thread.Sleep(2000);
 4             Console.WriteLine("Task 1 is cost 2 sec");
 5             throw new Exception("Exception in task 1");
 6         }
 7         public void Run2()
 8         {
 9             Thread.Sleep(3000);
10             Console.WriteLine("Task 2 is cost 3 sec");
11             throw new Exception("Exception in task 2");
12         }
13         public void Run3()
14         {
15             Thread.Sleep(5000);
16             Console.WriteLine("Task 3 is cost 5 sec");
17             throw new Exception("Exception in task 3");
18         }
19 
20         public void ParallelInvokeMethod()
21         {
22             stopWatch.Start();
23             try
24             {
25                 Parallel.Invoke(Run1, Run2, Run3);
26             }
27             catch (AggregateException ex)
28             {
29                 foreach (var item in ex.InnerExceptions)
30                 {
31                     Console.WriteLine(item.Message);
32                 }
33             }
34             stopWatch.Stop();
35             Console.WriteLine("Parallel run " + stopWatch.ElapsedMilliseconds + " ms.");
36 
37             stopWatch.Restart();
38             try
39             {
40                 Run1();
41                 Run2();
42                 Run3();
43             }
44             catch (Exception ex)
45             {
46                 Console.WriteLine(ex.Message);
47             }
48             stopWatch.Stop();
49             Console.WriteLine("Normal run " + stopWatch.ElapsedMilliseconds + " ms.");
50         }

Parallel.Invoke中捕捉每個方法的異常，串列方法能捕捉到Run1的異常。