Use FileStream overload that preallocates file on WriteAllBytes #51298
Conversation
|
Tagging subscribers to this area: @carlossanlop Issue DetailsContributes to #51116 Perf results obtained from the benchmarks on dotnet/performance#1774: BenchmarkDotNet=v0.12.1.1521-nightly, OS=Windows 10.0.19042.868 (20H2/October2020Update)
Intel Core i7-9750H CPU 2.60GHz, 1 CPU, 12 logical and 6 physical cores
.NET SDK=6.0.100-preview.3.21207.34
[Host] : .NET 6.0.0 (6.0.21.20104), X64 RyuJIT
Job-EZBBOL : .NET 6.0.0 (42.42.42.42424), X64 RyuJIT
Job-BXKKVU : .NET 6.0.0 (42.42.42.42424), X64 RyuJIT
PowerPlanMode=00000000-0000-0000-0000-000000000000 Arguments=/p:DebugType=portable IterationTime=250.0000 ms
MaxIterationCount=101 MinIterationCount=100 WarmupCount=1
| Method | Toolchain | userBuffer | Mean | Error | StdDev | Median | Min | Max | Ratio | Allocated |
|------------------- |---------- |----------- |-------------:|-------------:|-------------:|-------------:|-------------:|-------------:|------:|----------:|
| WriteAllBytes |base | 512 | 358.6 μs | 8.24 μs | 24.44 μs | 359.5 μs | 286.6 μs | 416.3 μs | 1.00 | 4,355 B |
| WriteAllBytes |new | 512 | 328.8 μs | 5.02 μs | 14.66 μs | 329.8 μs | 277.6 μs | 354.4 μs | 0.92 | 4,472 B |
| | | | | | | | | | | |
| WriteAllBytesAsync |base | 512 | 424.5 μs | 5.59 μs | 16.21 μs | 423.7 μs | 392.0 μs | 468.1 μs | 1.00 | 5,184 B |
| WriteAllBytesAsync |new | 512 | 404.0 μs | 4.23 μs | 12.41 μs | 404.4 μs | 378.7 μs | 435.9 μs | 0.95 | 5,312 B |
| | | | | | | | | | | |
| WriteAllBytes |base | 4096 | 360.7 μs | 9.54 μs | 27.98 μs | 361.9 μs | 278.2 μs | 435.0 μs | 1.00 | 224 B |
| WriteAllBytes |new | 4096 | 321.6 μs | 7.09 μs | 20.56 μs | 328.2 μs | 267.1 μs | 351.0 μs | 0.90 | 352 B |
| | | | | | | | | | | |
| WriteAllBytesAsync |base | 4096 | 401.0 μs | 5.81 μs | 16.58 μs | 401.3 μs | 368.1 μs | 441.7 μs | 1.00 | 5,184 B |
| WriteAllBytesAsync |new | 4096 | 414.3 μs | 7.14 μs | 19.91 μs | 410.8 μs | 369.5 μs | 473.8 μs | 1.03 | 5,312 B |
| | | | | | | | | | | |
| WriteAllBytes |base | 16384 | 1,634.3 μs | 90.53 μs | 266.94 μs | 1,666.7 μs | 857.9 μs | 2,257.5 μs | 1.00 | 225 B |
| WriteAllBytes |new | 16384 | 1,397.7 μs | 105.00 μs | 311.24 μs | 1,339.7 μs | 824.2 μs | 1,994.4 μs | 0.88 | 353 B |
| | | | | | | | | | | |
| WriteAllBytesAsync |base | 16384 | 1,643.1 μs | 103.07 μs | 305.51 μs | 1,651.1 μs | 1,043.2 μs | 2,249.3 μs | 1.00 | 905 B |
| WriteAllBytesAsync |new | 16384 | 1,732.3 μs | 121.13 μs | 359.06 μs | 1,803.8 μs | 1,077.1 μs | 2,361.7 μs | 1.10 | 1,033 B |
| | | | | | | | | | | |
| WriteAllBytes |base | 1048576 | 1,276.9 μs | 53.03 μs | 156.37 μs | 1,303.9 μs | 832.7 μs | 1,629.7 μs | 1.00 | 225 B |
| WriteAllBytes |new | 1048576 | 1,393.0 μs | 49.60 μs | 147.03 μs | 1,379.7 μs | 1,088.6 μs | 1,666.0 μs | 1.11 | 353 B |
| | | | | | | | | | | |
| WriteAllBytesAsync |base | 1048576 | 2,388.1 μs | 51.55 μs | 152.81 μs | 2,418.3 μs | 1,825.8 μs | 2,670.0 μs | 1.00 | 905 B |
| WriteAllBytesAsync |new | 1048576 | 2,234.9 μs | 71.46 μs | 211.83 μs | 2,284.9 μs | 1,287.2 μs | 2,580.5 μs | 0.94 | 1,033 B |
| | | | | | | | | | | |
| WriteAllBytes |base | 104857600 | 177,801.0 μs | 14,731.29 μs | 43,665.55 μs | 188,410.9 μs | 89,479.8 μs | 264,803.5 μs | 1.00 | 368 B |
| WriteAllBytes |new | 104857600 | 181,757.1 μs | 10,989.27 μs | 32,573.70 μs | 179,412.4 μs | 92,253.6 μs | 243,485.6 μs | 1.11 | 424 B |
| | | | | | | | | | | |
| WriteAllBytesAsync |base | 104857600 | 184,642.8 μs | 15,042.38 μs | 44,587.66 μs | 193,423.9 μs | 108,295.1 μs | 261,851.5 μs | 1.00 | 1,048 B |
| WriteAllBytesAsync |new | 104857600 | 175,904.1 μs | 16,981.90 μs | 50,336.65 μs | 187,221.0 μs | 82,085.9 μs | 259,399.6 μs | 1.03 | 1,176 B |
|
|
Note regarding the This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, to please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change. |
|
@jozkee it looks like we are not going to get any speedup in this case. Could find a scenario that uses a different pattern: multiple calls to |
|
Closing this since it proves that is not beneficial to preallocate a file if we only perform one single Write.
@adamsitnik yup, I will measure the benefits of no buffering on |
ghost
locked as resolved and limited conversation to collaborators
Contributes to #51116
Simple change on WriteAllBytes[Async] in order to verify if there is a perf benefit on using the new FileStream ctor. This builds on top of #51111.
Perf results obtained from the benchmarks on dotnet/performance#1774:
Note: allocations increased by 128 B and appears that large writes (>=16Kb) show a perf regression of 10%. Also note that min/maxIterationCount is 100 - 101 in order to neglect the abnormal results.