PipeThroughputBenchmark

Enviorment

SET COMPlus_TieredCompilation=0

PipeThroughputBenchmark.cs

using System;
using System.Buffers;
using System.IO.Pipelines;
using System.Threading.Tasks;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;
using Microsoft.AspNetCore.Server.Kestrel.Transport.Abstractions.Internal;

public class PipeThroughputBenchmark
{
    // 128 (min Length) * InnerLoopCount > _memoryPool.MaxBufferSize
    private const int InnerLoopCount = 33;

    private PipeReader _reader;
    private PipeWriter _writer;
    private MemoryPool<byte> _memoryPool;

    [GlobalSetup]
    public void Setup()
    {
        _memoryPool = KestrelMemoryPool.Create();

        var chunkLength = Length / Chunks;
        if (chunkLength > _memoryPool.MaxBufferSize)
        {
            // Parallel test will deadlock if too large (waiting for second Task to complete), so N/A that run
            throw new InvalidOperationException();
        }

        if (Length != chunkLength * Chunks)
        {
            // Test will deadlock waiting for data so N/A that run
            throw new InvalidOperationException();
        }

        var pipe = new Pipe(new PipeOptions(_memoryPool));
        _reader = pipe.Reader;
        _writer = pipe.Writer;
    }

    [Params(1, 2, 4, 8)]
    public int Chunks { get; set; }

    [Params(128, 256, 1024, 2048, 4096, 8192)]
    public int Length { get; set; }

    [Benchmark(OperationsPerInvoke = InnerLoopCount)]
    public Task Parse_ParallelAsync()
    {
        return Parse_ParallelAsyncImpl();
    }

    private Task Parse_ParallelAsyncImpl()
    {
        var writing = Task.Run(async () =>
        {
            var chunks = Chunks;
            var chunkLength = Length / chunks;
            for (int i = 0; i < InnerLoopCount; i++)
            {
                for (var c = 0; c < chunks; c++)
                {
                    _writer.GetMemory(chunkLength);
                    _writer.Advance(chunkLength);
                    await _writer.FlushAsync();
                }
            }
        });

        var reading = Task.Run(async () =>
        {
            long remaining = InnerLoopCount * Length;
            while (remaining != 0)
            {
                var result = await _reader.ReadAsync();
                remaining -= result.Buffer.Length;
                _reader.AdvanceTo(result.Buffer.End, result.Buffer.End);
            }
        });

        return Task.WhenAll(writing, reading);
    }

    [Benchmark(OperationsPerInvoke = InnerLoopCount)]
    public Task Parse_SequentialAsync()
    {
        return Parse_SequentialAsyncImpl();
    }

    private async Task Parse_SequentialAsyncImpl()
    {
        var chunks = Chunks;
        var chunkLength = Length / chunks;

        for (int i = 0; i < InnerLoopCount; i++)
        {
            for (var c = 0; c < chunks; c++)
            {
                _writer.GetMemory(chunkLength);
                _writer.Advance(chunkLength);
                await _writer.FlushAsync();
            }

            var result = await _reader.ReadAsync();
            _reader.AdvanceTo(result.Buffer.End, result.Buffer.End);
        }
    }

    [GlobalCleanup]
    public void Cleanup()
    {
        _memoryPool.Dispose();
    }

    public static void Main(string[] args)
    {
        var summary = BenchmarkSwitcher.FromAssembly(typeof(PipeThroughputBenchmark).Assembly).Run(args);
    }
}

PipeThroughputBenchmark.csproj

<Project Sdk="Microsoft.NET.Sdk">

  <PropertyGroup>
    <OutputType>Exe</OutputType>
    <TargetFramework>netcoreapp3.0</TargetFramework>
  </PropertyGroup>

  <ItemGroup>
    <PackageReference Include="BenchmarkDotNet" Version="0.11.3" />
    <PackageReference Include="Microsoft.AspNetCore.Server.Kestrel" Version="2.2.0" />
    <PackageReference Include="System.IO.Pipelines" Version="4.5.2" />
  </ItemGroup>

</Project>

                Method | Chunks | Length |       Mean |      Error |     StdDev |
---------------------- |------- |------- |-----------:|-----------:|-----------:|
   Parse_ParallelAsync |      1 |    128 |   615.0 ns |  3.1738 ns |  2.9688 ns |
 Parse_SequentialAsync |      1 |    128 |   322.6 ns |  0.5409 ns |  0.4795 ns |
   Parse_ParallelAsync |      1 |    256 |   605.8 ns |  1.6088 ns |  1.5048 ns |
 Parse_SequentialAsync |      1 |    256 |   321.8 ns |  0.6478 ns |  0.5743 ns |
   Parse_ParallelAsync |      1 |   1024 |   663.3 ns |  1.6789 ns |  1.5705 ns |
 Parse_SequentialAsync |      1 |   1024 |   322.2 ns |  0.7686 ns |  0.7190 ns |
   Parse_ParallelAsync |      1 |   2048 |   809.4 ns |  4.0722 ns |  3.8091 ns |
 Parse_SequentialAsync |      1 |   2048 |   323.3 ns |  1.5330 ns |  1.4340 ns |
   Parse_ParallelAsync |      1 |   4096 | 1,043.7 ns |  3.1687 ns |  2.9640 ns |
 Parse_SequentialAsync |      1 |   4096 |   329.6 ns |  0.5404 ns |  0.4512 ns |
   Parse_ParallelAsync |      1 |   8192 |         NA |         NA |         NA |
 Parse_SequentialAsync |      1 |   8192 |         NA |         NA |         NA |
   Parse_ParallelAsync |      2 |    128 | 1,043.0 ns |  2.2123 ns |  1.9611 ns |
 Parse_SequentialAsync |      2 |    128 |   436.3 ns |  2.3291 ns |  2.1786 ns |
   Parse_ParallelAsync |      2 |    256 | 1,033.3 ns |  4.1923 ns |  3.9215 ns |
 Parse_SequentialAsync |      2 |    256 |   437.4 ns | 10.8902 ns | 10.1867 ns |
   Parse_ParallelAsync |      2 |   1024 | 1,064.3 ns |  2.8559 ns |  2.5317 ns |
 Parse_SequentialAsync |      2 |   1024 |   434.6 ns |  1.8056 ns |  1.5078 ns |
   Parse_ParallelAsync |      2 |   2048 | 1,136.8 ns |  3.4331 ns |  3.2113 ns |
 Parse_SequentialAsync |      2 |   2048 |   428.5 ns |  2.2958 ns |  2.0352 ns |
   Parse_ParallelAsync |      2 |   4096 | 1,306.6 ns |  4.1388 ns |  3.8714 ns |
 Parse_SequentialAsync |      2 |   4096 |   434.1 ns |  1.7415 ns |  1.4542 ns |
   Parse_ParallelAsync |      2 |   8192 | 1,659.6 ns |  3.9254 ns |  3.6718 ns |
 Parse_SequentialAsync |      2 |   8192 |   498.7 ns |  1.3643 ns |  1.2094 ns |
   Parse_ParallelAsync |      4 |    128 | 1,658.4 ns | 10.6838 ns |  9.4709 ns |
 Parse_SequentialAsync |      4 |    128 |   670.2 ns |  1.8434 ns |  1.6341 ns |
   Parse_ParallelAsync |      4 |    256 | 1,644.4 ns |  6.7991 ns |  5.3083 ns |
 Parse_SequentialAsync |      4 |    256 |   692.3 ns | 13.7884 ns | 12.8977 ns |
   Parse_ParallelAsync |      4 |   1024 | 1,771.7 ns | 33.7018 ns | 36.0606 ns |
 Parse_SequentialAsync |      4 |   1024 |   653.0 ns |  2.6855 ns |  2.0966 ns |
   Parse_ParallelAsync |      4 |   2048 | 1,721.1 ns | 21.0045 ns | 19.6476 ns |
 Parse_SequentialAsync |      4 |   2048 |   665.9 ns |  2.1371 ns |  1.6685 ns |
   Parse_ParallelAsync |      4 |   4096 | 1,872.4 ns | 13.2456 ns | 12.3899 ns |
 Parse_SequentialAsync |      4 |   4096 |   649.4 ns |  2.7504 ns |  2.5727 ns |
   Parse_ParallelAsync |      4 |   8192 | 2,189.3 ns |  5.1524 ns |  4.8195 ns |
 Parse_SequentialAsync |      4 |   8192 |   728.4 ns |  4.1580 ns |  3.8894 ns |
   Parse_ParallelAsync |      8 |    128 | 2,797.3 ns | 16.9029 ns | 15.8110 ns |
 Parse_SequentialAsync |      8 |    128 | 1,093.9 ns |  5.8144 ns |  5.4388 ns |
   Parse_ParallelAsync |      8 |    256 | 2,767.8 ns | 27.9809 ns | 26.1734 ns |
 Parse_SequentialAsync |      8 |    256 | 1,075.9 ns |  5.6136 ns |  5.2510 ns |
   Parse_ParallelAsync |      8 |   1024 | 2,805.9 ns | 73.2250 ns | 71.9168 ns |
 Parse_SequentialAsync |      8 |   1024 | 1,089.7 ns |  2.8515 ns |  2.6673 ns |
   Parse_ParallelAsync |      8 |   2048 | 2,822.3 ns | 21.9266 ns | 20.5101 ns |
 Parse_SequentialAsync |      8 |   2048 | 1,082.8 ns |  7.1544 ns |  6.6922 ns |
   Parse_ParallelAsync |      8 |   4096 | 3,052.4 ns | 29.3138 ns | 27.4201 ns |
 Parse_SequentialAsync |      8 |   4096 | 1,098.1 ns |  8.8584 ns |  8.2861 ns |
   Parse_ParallelAsync |      8 |   8192 | 3,335.2 ns | 30.3906 ns | 28.4274 ns |
 Parse_SequentialAsync |      8 |   8192 | 1,157.9 ns |  1.7976 ns |  1.5935 ns |