C# 多线程 Parallel.ForEach 和 ForEach 效率问题研究及理解

最近要做一个大数据dataTable循环操作，开始发现运用foreach，进行大数据循环，并做了一些逻辑处理。在循环中耗费的时间过长。后来换成使用Parallel.ForEach来进行循环。一开始认为，数据比较大时，Parallel.ForEach肯定比 ForEach效率高，后来发现，其实并不是这样。

我用了1000万次循环测试：

using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Linq;
using System.Text;
using System.Threading.Tasks;

namespace ConsoleApplication1
{
    class Program
    {
        static void Main(string[] args)
        {
         
            Stopwatch Watch1 = new Stopwatch();
            Watch1.Start();
            List<entityA> source = new List<entityA>();
            for (int i = 0; i < 10000000; i++)
            {
                source.Add(new entityA
                {
                    name = "悟空" + i,
                    sex = i % 2 == 0 ? "男" : "女",
                    age = i
                });
            }
            Watch1.Stop();

            Console.WriteLine("list循环插入耗时：" + Watch1.ElapsedMilliseconds);
            Stopwatch Watch2 = new Stopwatch();
            Watch2.Start();
            loop1(source);
            Watch2.Stop();
            Console.WriteLine("一般for循环耗时：" + Watch2.ElapsedMilliseconds);

            Stopwatch Watch3 = new Stopwatch();
            Watch3.Start();
            loop2(source);
            Watch3.Stop();
            Console.WriteLine("一般foreach循环耗时：" + Watch3.ElapsedMilliseconds);


            Stopwatch Watch4 = new Stopwatch();
            Watch4.Start();
            loop3(source);
            Watch4.Stop();
            Console.WriteLine("并行for循环耗时：" + Watch4.ElapsedMilliseconds);

            Stopwatch Watch5 = new Stopwatch();
            Watch5.Start();
            loop4(source);
            Watch5.Stop();
            Console.WriteLine("并行foreach循环耗时：" + Watch5.ElapsedMilliseconds);
            Console.ReadLine();
        }
        //普通的for循环
        static void loop1(List<entityA> source)
        {
            int count = source.Count();
            for (int i = 0; i < count; i++)
            {
                source[0].age= + 10;
                //System.Threading.Thread.Sleep(10);
            }
        }

        //普通的foreach循环
        static void loop2(List<entityA> source)
        {
            foreach (entityA item in source)
            {
                item.age =+ 10;
                //System.Threading.Thread.Sleep(10);
            }
        }

        //并行的for循环
        static void loop3(List<entityA> source)
        {
            int count = source.Count();
            Parallel.For(0, count, item =>
            {
                //source[count].age= source[count].age + 10;
                //System.Threading.Thread.Sleep(10);
            });
        }

        //并行的foreach循环
        static void loop4(List<entityA> source)
        {
            Parallel.ForEach(source, item =>
            {
                item.age = item.age + 10;
                //System.Threading.Thread.Sleep(10);
            });
        }
    }


    //简单的实体
    class entityA
    {
        public string name { set; get; }
        public string sex { set; get; }
        public int age { set; get; }
    }
}

运行结果：

结果居然是并行比一般的循环还耗时，但这是为什么呢？

这是因为循环体内执行的任务开销太小，仅仅是age+10 而已。微软的文章已经指出任务的开销大小对并行任务的影响。如果任务很小，那么由于并行管理的附加开销（任务分配，调度，同步等成本），可能并行执行并不是优化方案。这也是上述程序Foreach与For效率高出的原因。

基于这一点，我们对程序进行调整，循环1000次，每次里面线程sleep(10),这样我们试试。

using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Linq;
using System.Text;
using System.Threading.Tasks;

namespace ConsoleApplication1
{
    class Program
    {
        static void Main(string[] args)
        {
         
            Stopwatch Watch1 = new Stopwatch();
            Watch1.Start();
            List<entityA> source = new List<entityA>();
            for (int i = 0; i < 1000; i++)
            {
                source.Add(new entityA
                {
                    name = "悟空" + i,
                    sex = i % 2 == 0 ? "男" : "女",
                    age = i
                });
            }
            Watch1.Stop();

            Console.WriteLine("list循环插入耗时：" + Watch1.ElapsedMilliseconds);
            Stopwatch Watch2 = new Stopwatch();
            Watch2.Start();
            loop1(source);
            Watch2.Stop();
            Console.WriteLine("一般for循环耗时：" + Watch2.ElapsedMilliseconds);

            Stopwatch Watch3 = new Stopwatch();
            Watch3.Start();
            loop2(source);
            Watch3.Stop();
            Console.WriteLine("一般foreach循环耗时：" + Watch3.ElapsedMilliseconds);


            Stopwatch Watch4 = new Stopwatch();
            Watch4.Start();
            loop3(source);
            Watch4.Stop();
            Console.WriteLine("并行for循环耗时：" + Watch4.ElapsedMilliseconds);

            Stopwatch Watch5 = new Stopwatch();
            Watch5.Start();
            loop4(source);
            Watch5.Stop();
            Console.WriteLine("并行foreach循环耗时：" + Watch5.ElapsedMilliseconds);
            Console.ReadLine();
        }
        //普通的for循环
        static void loop1(List<entityA> source)
        {
            int count = source.Count();
            for (int i = 0; i < count; i++)
            {
                source[0].age= + 10;
                System.Threading.Thread.Sleep(10);
            }
        }

        //普通的foreach循环
        static void loop2(List<entityA> source)
        {
            foreach (entityA item in source)
            {
                item.age =+ 10;
                System.Threading.Thread.Sleep(10);
            }
        }

        //并行的for循环
        static void loop3(List<entityA> source)
        {
            int count = source.Count();
            Parallel.For(0, count, item =>
            {
                //source[count].age= source[count].age + 10;
                System.Threading.Thread.Sleep(10);
            });
        }

        //并行的foreach循环
        static void loop4(List<entityA> source)
        {
            Parallel.ForEach(source, item =>
            {
                item.age = item.age + 10;
                System.Threading.Thread.Sleep(10);
            });
        }
    }


    //简单的实体
    class entityA
    {
        public string name { set; get; }
        public string sex { set; get; }
        public int age { set; get; }
    }
}

执行结果：