System.Threading.Tasks.Parallel命名空间下的静态类Parallel用来代替Task类
一、Parallel.Invoke:
试图将很多方法并行运行,如果传入的是4个方法,则至少需要4个逻辑内核才能足以让这4个方法并发运行,逻辑内核也称为硬件线程。、
需要注意的是:1.即使拥有4个逻辑内核,也不一定能够保证所需要运行的4个方法能够同时启动运行,如果其中的一个内核处于繁忙状态,那么底层的调度逻辑可能会延迟某些方法的初始化执行。
2.通过Parallel.Invoke编写的并发执行代码一定不能依赖与特定的执行顺序,因为它的并发执行顺序也是不定的。
3.使用Parallel.Invoke方法一定要测量运行结果、实现加速比以及逻辑内核的使用率,这点很重要。
4.使用Parallel.Invoke,在运行并行方法前都会产生一些额外的开销,如分配硬件线程等。
好处:这是一种并行运行很多方法的简单方式,使用Parallel.Invoke,不需要考虑任务和线程的问题。
class Program
{
private static List<Product> ProductList = null;
/* 没有特定的执行顺序
* 示例中 基于电脑配置 采用了4个方法的并行编程
* Parallel.Invoke 首先会尝试并行启动4个方法,充分利用一个或多个物理处理器所提供的多个逻辑内核
* 但是在实际的并行执行中,至少要有4个逻辑内核才能满足4个方法的并行运行
* 如果有个或者多个逻辑内核处于繁忙状态,那么底层的调度逻辑可能会延迟某些方法的初始化执行
* 通过Parallel.Invoke编写的并发执行代码一定不能依赖与特定的执行顺序,因为它的并发执行顺序也是不定的。
*/
static void Main(string[] args)
{
ProductList = new List<Product>();
Thread.Sleep(3000);
Stopwatch swTask = new Stopwatch();
swTask.Start();
/*执行并行操作*/
Parallel.Invoke(SetProcuct1_500, SetProcuct2_500, SetProcuct3_500, SetProcuct4_500);
swTask.Stop();
Console.WriteLine("500条数据 并行编程所耗时间:" + swTask.ElapsedMilliseconds);
ProductList = new List<Product>();
Thread.Sleep(3000);/*防止并行操作 与 顺序操作冲突*/
Stopwatch sw = new Stopwatch();
sw.Start();
SetProcuct1_500();
SetProcuct2_500();
SetProcuct3_500();
SetProcuct4_500();
sw.Stop();
Console.WriteLine("500条数据 顺序编程所耗时间:" + sw.ElapsedMilliseconds);
ProductList = new List<Product>();
Thread.Sleep(3000);
swTask.Restart();
/*执行并行操作*/
Parallel.Invoke(() => SetProcuct1_10000(), () => SetProcuct2_10000(), () => SetProcuct3_10000(), () => SetProcuct4_10000());
swTask.Stop();
Console.WriteLine("10000条数据 并行编程所耗时间:" + swTask.ElapsedMilliseconds);
ProductList = new List<Product>();
Thread.Sleep(3000);
sw.Restart();
SetProcuct1_10000();
SetProcuct2_10000();
SetProcuct3_10000();
SetProcuct4_10000();
sw.Stop();
Console.WriteLine("10000条数据 顺序编程所耗时间:" + sw.ElapsedMilliseconds);
Console.ReadLine();
}
private static void SetProcuct1_500()
{
for (int index = 1; index < 500; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct1 执行完成");
}
private static void SetProcuct2_500()
{
for (int index = 500; index < 1000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct2 执行完成");
}
private static void SetProcuct3_500()
{
for (int index = 1000; index < 2000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct3 执行完成");
}
private static void SetProcuct4_500()
{
for (int index = 2000; index < 3000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct4 执行完成");
}
private static void SetProcuct1_10000()
{
for (int index = 1; index < 20000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct1 执行完成");
}
private static void SetProcuct2_10000()
{
for (int index = 20000; index < 40000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct2 执行完成");
}
private static void SetProcuct3_10000()
{
for (int index = 40000; index < 60000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct3 执行完成");
}
private static void SetProcuct4_10000()
{
for (int index = 60000; index < 80000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct4 执行完成");
}
}
class Product
{
public string Name { get; set; }
public string Category { get; set; }
public int SellPrice { get; set; }
}
![](http://images.cnitblog.com/i/333725/201408/201730187376833.png)
图中我们可以看出利用 Parallel.Invoke编写的并发执行代,它的并发执行顺序也是不定的。
但是所执行的时间上比不采用并行编程所耗的时间差不多。
这是因为我们在并行编程中操作了共享资源 ProductList ,如果我把代码做出以下修改,采用并行编程的好处就显现出来了。
class Program
{
/* 没有特定的执行顺序
* 示例中 基于电脑配置 采用了4个方法的并行编程
* Parallel.Invoke 首先会尝试并行启动4个方法,充分利用一个或多个物理处理器所提供的多个逻辑内核
* 但是在实际的并行执行中,至少要有4个逻辑内核才能满足4个方法的并行运行
* 如果有个或者多个逻辑内核处于繁忙状态,那么底层的调度逻辑可能会延迟某些方法的初始化执行
* 通过Parallel.Invoke编写的并发执行代码一定不能依赖与特定的执行顺序,因为它的并发执行顺序也是不定的。
*/
static void Main(string[] args)
{
Thread.Sleep(3000);
Stopwatch swTask = new Stopwatch();
swTask.Start();
/*执行并行操作*/
Parallel.Invoke(SetProcuct1_500, SetProcuct2_500, SetProcuct3_500, SetProcuct4_500);
swTask.Stop();
Console.WriteLine("500条数据 并行编程所耗时间:" + swTask.ElapsedMilliseconds);
Thread.Sleep(3000);/*防止并行操作 与 顺序操作冲突*/
Stopwatch sw = new Stopwatch();
sw.Start();
SetProcuct1_500();
SetProcuct2_500();
SetProcuct3_500();
SetProcuct4_500();
sw.Stop();
Console.WriteLine("500条数据 顺序编程所耗时间:" + sw.ElapsedMilliseconds);
Thread.Sleep(3000);
swTask.Restart();
/*执行并行操作*/
Parallel.Invoke(() => SetProcuct1_10000(), () => SetProcuct2_10000(), () => SetProcuct3_10000(), () => SetProcuct4_10000());
swTask.Stop();
Console.WriteLine("10000条数据 并行编程所耗时间:" + swTask.ElapsedMilliseconds);
Thread.Sleep(3000);
sw.Restart();
SetProcuct1_10000();
SetProcuct2_10000();
SetProcuct3_10000();
SetProcuct4_10000();
sw.Stop();
Console.WriteLine("10000条数据 顺序编程所耗时间:" + sw.ElapsedMilliseconds);
Console.ReadLine();
}
private static void SetProcuct1_500()
{
List<Product> ProductList = new List<Product>();
for (int index = 1; index < 500; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct1 执行完成");
}
private static void SetProcuct2_500()
{
List<Product> ProductList = new List<Product>();
for (int index = 500; index < 1000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct2 执行完成");
}
private static void SetProcuct3_500()
{
List<Product> ProductList = new List<Product>();
for (int index = 1000; index < 2000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct3 执行完成");
}
private static void SetProcuct4_500()
{
List<Product> ProductList = new List<Product>();
for (int index = 2000; index < 3000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct4 执行完成");
}
private static void SetProcuct1_10000()
{
List<Product> ProductList = new List<Product>();
for (int index = 1; index < 20000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct1 执行完成");
}
private static void SetProcuct2_10000()
{
List<Product> ProductList = new List<Product>();
for (int index = 20000; index < 40000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct2 执行完成");
}
private static void SetProcuct3_10000()
{
List<Product> ProductList = new List<Product>();
for (int index = 40000; index < 60000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct3 执行完成");
}
private static void SetProcuct4_10000()
{
List<Product> ProductList = new List<Product>();
for (int index = 60000; index < 80000; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
}
Console.WriteLine("SetProcuct4 执行完成");
}
}
class Product
{
public string Name { get; set; }
public string Category { get; set; }
public int SellPrice { get; set; }
}
我将每个方法中的资源隔离,性能显而易见。
但是在操作500条数据时,显然采用并行操作并不明智,并行所带来的损耗比较大,在实际的开发中,还是要注意下是否有必要进行并行编程。
二、Parallel.For
将for循环替换成Parallel.For,并采用适合这个新方法的参数,就可以对这个已有的for循环进行重构,使其能够充分利用并行化优势
需要注意的是:1.Parallel.For不支持浮点数的步进,使用的是Int32或Int64,每一次迭代的时候加1
2.由于循环体是并行运行,去迭代执行的顺序无法保证
class Program
{
static void Main(string[] args)
{
Thread.Sleep(3000);
ForSetProcuct_100();
Thread.Sleep(3000);
ParallelForSetProcuct_100();
Console.ReadLine();
}
private static void ForSetProcuct_100()
{
Stopwatch sw = new Stopwatch();
sw.Start();
List<Product> ProductList = new List<Product>();
for (int index = 1; index < 100; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
Console.WriteLine("for SetProcuct index: {0}", index);
}
sw.Stop();
Console.WriteLine("for SetProcuct 10 执行完成 耗时:{0}", sw.ElapsedMilliseconds);
}
private static void ParallelForSetProcuct_100()
{
Stopwatch sw = new Stopwatch();
sw.Start();
List<Product> ProductList = new List<Product>();
Parallel.For(1, 100, index =>
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
ProductList.Add(model);
Console.WriteLine("ForSetProcuct SetProcuct index: {0}", index);
});
sw.Stop();
Console.WriteLine("ForSetProcuct SetProcuct 20000 执行完成 耗时:{0}", sw.ElapsedMilliseconds);
}
}
class Product
{
public string Name { get; set; }
public string Category { get; set; }
public int SellPrice { get; set; }
}
![](http://images.cnitblog.com/i/333725/201408/211103175656221.png)
三、Parallel.ForEach
Parallel.ForEach提供一个并行处理一组数据的机制,可以利用一个范围的整数作为一组数据,然后通过一个自定义的分区器将这个范围转换为一组数据块,每一块数据都通过循环的方式进行处理,而这些循环式并行执行的。
class Program
{
static void Main(string[] args)
{
List<Product> ProductList =GetProcuctList();
Parallel.ForEach(ProductList, (model) => {
Console.WriteLine(model.Name);
});
Console.ReadLine();
}
private static List<Product> GetProcuctList()
{
List<Product> result = new List<Product>();
for (int index = 1; index < 100; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
result.Add(model);
}
return result;
}
}
class Product
{
public string Name { get; set; }
public string Category { get; set; }
public int SellPrice { get; set; }
}
![](http://images.cnitblog.com/i/333725/201408/211114122682437.png)
四、ParallelLoopState
ParallelLoopState该实例提供了以下两个方法用于停止 Parallel.For,Parallel.ForEach
Break-这个方法告诉并行循环应该在执行了当前迭代后尽快地停止执行。如果调用Break时正在处理迭代100,那么循环仍然会处理所有小于100的迭代。
Stop-这个方法告诉并行循环应该尽快停止执行,如果调用Stop时迭代100正在被处理,那么循环无法保证处理完所有小于100的迭代
class Program
{
static void Main(string[] args)
{
List<Product> productList = GetProcuctList_500();
Thread.Sleep(3000);
Parallel.For(0, productList.Count, (i, loopState) =>
{
if (i < 100)
{
Console.WriteLine("采用Stop index:{0}", i);
}
else
{
/* 满足条件后 尽快停止执行,无法保证小于100的索引数据全部输出*/
loopState.Stop();
return;
}
});
Thread.Sleep(3000);
Parallel.For(0, productList.Count, (i, loopState) =>
{
if (i < 100)
{
Console.WriteLine("采用Break index:{0}", i);
}
else
{
/* 满足条件后 尽快停止执行,保证小于100的索引数据全部输出*/
loopState.Break();
return;
}
});
Thread.Sleep(3000);
Parallel.ForEach(productList, (model, loopState) =>
{
if (model.SellPrice < 10)
{
Console.WriteLine("采用Stop index:{0}", model.SellPrice);
}
else
{
/* 满足条件后 尽快停止执行,无法保证满足条件的数据全部输出*/
loopState.Stop();
return;
}
});
Thread.Sleep(3000);
Parallel.ForEach(productList, (model, loopState) =>
{
if (model.SellPrice < 10)
{
Console.WriteLine("采用Break index:{0}", model.SellPrice);
}
else
{
/* 满足条件后 尽快停止执行,保证满足条件的数据全部输出*/
loopState.Break();
return;
}
});
Console.ReadLine();
}
private static List<Product> GetProcuctList_500()
{
List<Product> result = new List<Product>();
for (int index = 1; index < 500; index++)
{
Product model = new Product();
model.Category = "Category" + index;
model.Name = "Name" + index;
model.SellPrice = index;
result.Add(model);
}
return result;
}
}
![](http://images.cnitblog.com/blog/333725/201408/211512423628251.png)
由图中可以看出Break可以保证输出满足所有条件的数据,而Stop则无法保证。