ConcurrentBag 与 List 的性能比较
本文关键字:性能比 比较 性能 List ConcurrentBag | 更新日期: 2023-09-27 18:34:44
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:
袋00:00:00.0144421
叠00:00:00.0341379
问00:00:00.0400114
列表删除 000:00:00.6188329
列表在末尾删除00:00:00.6202170