ConcurrentBag 与 List 的性能比较



前言:我问这个只是因为我没有一个环境(数据集足够大+计算能力(来以可靠的方式测试它。

问题:给定一个ConcurrentBag<T>,加载了数十亿个项目,由单个线程访问/使用,它的性能是否类似于List<T>?换句话说,对ConcurrentBag<T>的枚举是否比对List<T>的枚举具有或更差的性能?

ConcurrentBag<T>的性能

将不可避免地低于List<T>。尽管您只能从单个线程访问它,但该结构仍然需要具有适当的机制,以防止在出现并发访问时出现竞争危害的可能性。

如果要在开始枚举之前从单个线程加载集合,则可以通过使用 ConcurrentBag(IEnumerable<T>) 构造函数来避免性能开销,而不是通过其 Add 方法单独添加每个项。

ConcurrentBag<T>为枚举提供"时刻快照"语义;请参阅其GetEnumerator方法的备注。当您从 foreach 循环访问ConcurrentBag<T>时,它会首先将其全部内容复制到普通List<T>中,然后枚举。每次在循环中使用它时,这将产生大量的性能开销(计算和内存方面(。

如果你的方案是你的列表将由多个线程填充,但随后只由一个线程读取,那么你应该在编写器完成后立即将其转换为List<T>

数十亿个项目和列表或并发包?这是"不行"。

就性能而言,请尝试测试添加:(随意修改此内容以测试其他操作(

using System;
using System.Collections.Concurrent;
using System.Collections.Generic;
using System.Diagnostics;
using System.Linq;
using System.Text;
using System.Threading;
using System.Threading.Tasks;
namespace ConcurrentBagTest
{
    // You must compile this for x64 or you will get OutOfMemory exception
    class Program
    {
        static void Main(string[] args)
        {
            ListTest(10000000);
            ListTest(100000000);
            ListTest(1000000000);
            ConcurrentBagTest(10000000);
            ConcurrentBagTest(100000000);
            Console.ReadKey();
        }

        static void ConcurrentBagTest(long count)
        {
            try
            {
                var bag = new ConcurrentBag<long>();
                Console.WriteLine($"--- ConcurrentBagTest count = {count}");
                Console.WriteLine($"I will use {(count * sizeof(long)) / Math.Pow(1024, 2)} MiB of RAM");
                Stopwatch stopwatch = new Stopwatch();
                stopwatch.Start();
                for (long i = 0; i < count; i++)
                {
                    bag.Add(i);
                }
                stopwatch.Stop();
                Console.WriteLine($"Inserted {bag.LongCount()} items in {stopwatch.Elapsed.TotalSeconds} s");
                Console.WriteLine();
                Console.WriteLine();
            }
            catch (Exception ex)
            {
                Console.WriteLine(ex.ToString());
            }
            GC.Collect();
            GC.WaitForPendingFinalizers();
        }
        static void ListTest(long count)
        {
            try
            {
                var list = new List<long>();
                Console.WriteLine($"--- ListTest count = {count}");
                Console.WriteLine($"I will use {(count * sizeof(long)) / Math.Pow(1024, 2)} MiB of RAM");
                Stopwatch stopwatch = new Stopwatch();
                stopwatch.Start();
                for (long i = 0; i < count; i++)
                {
                    list.Add(i);
                }
                stopwatch.Stop();
                Console.WriteLine($"Inserted {list.LongCount()} items in {stopwatch.Elapsed.TotalSeconds} s");
                Console.WriteLine();
                Console.WriteLine();
            }
            catch (Exception ex)
            {
                Console.WriteLine(ex.ToString());
            }
            GC.Collect();
            GC.WaitForPendingFinalizers();
        }
    }
}

我的输出:

--- ListTest count = 10000000
I will use 76,2939453125 MiB of RAM
Inserted 10000000 items in 0,0807315 s

--- ListTest count = 100000000
I will use 762,939453125 MiB of RAM
Inserted 100000000 items in 0,7741546 s

--- ListTest count = 1000000000
I will use 7629,39453125 MiB of RAM
System.OutOfMemoryException: Array dimensions exceeded supported range.
--- ConcurrentBagTest count = 10000000
I will use 76,2939453125 MiB of RAM
Inserted 10000000 items in 1,0744069 s

--- ConcurrentBagTest count = 100000000
I will use 762,939453125 MiB of RAM
Inserted 100000000 items in 11,3976436 s

使用 CPU:英特尔酷睿 i7-2600 @ 3.4 GHz,

使用内存:16 GB

另请查看此答案以了解限制。

但是,如果您需要删除项目,ConcurrentBag 比 List 快得多

void Main()
{
    ConcurrentBag<int> bag = new ConcurrentBag<int>();
    ConcurrentStack<int> stack = new ConcurrentStack<int>();
    ConcurrentQueue<int> q = new ConcurrentQueue<int>();
    List<int> list = new List<int>();
    Stopwatch sw = new Stopwatch();
    int count = 100000;
    sw.Start();
    for (int i = 0; i < count; i++)
    {
        bag.Add(i);
    }
    for (int i = 0; i< count; i++)
    {
        bag.TryTake(out _);
    }
    sw.Elapsed.Dump("BAG");
    sw.Start();
    for (int i = 0; i < count; i++)
    {
        stack.Push(i);
    }
    for (int i = 0; i < count; i++)
    {
        stack.TryPop(out _);
    }
    sw.Elapsed.Dump("Stack");
    sw.Start();
    for (int i = 0; i < count; i++)
    {
        q.Enqueue(i);
    }
    for (int i = 0; i < count; i++)
    {
        q.TryDequeue(out _);
    }
    sw.Elapsed.Dump("Q");
    sw.Start();
    for (int i = 0; i < count; i++)
    {
        list.Add(i);
    }
    for (int i = 0; i < count; i++)
    {
        list.RemoveAt(0);
    }
    sw.Elapsed.Dump("list remove at 0");
    sw.Start();
    for (int i = 0; i < count; i++)
    {
        list.Add(i);
    }
    for (int i = 0; i < count; i++)
    {
        list.RemoveAt(list.Count -1);
    }
    sw.Elapsed.Dump("list remove at end");
}

结果:

00:00:00.0144421

00:00:00.0341379

00:00:00.0400114

列表删除 000:00:00.6188329

列表在末尾删除00:00:00.6202170

最新更新