| commit | f8dd838d8c7de318b0734b3f1afa78a9d976bf09 | [log] [tgz] |
|---|---|---|
| author | Matthias Frei <matthias.frei@inf.ethz.ch> | Sat May 01 19:46:39 2021 +0000 |
| committer | Tobias Klauser <tobias.klauser@gmail.com> | Sat May 01 22:26:12 2021 +0000 |
| tree | c2f9971046c93ba835e85d24fffdc340345aee91 | |
| parent | aec3718b3fa014837019136dd7561a23acb262fc [diff] |
internal/socket: reuse buffers in recv/sendMsgs Use a pool to reuse internally allocated temporary buffers, i.e. the mmsghdrs structures and the associated iovecs and sockaddr buffers. The temporary buffers obtained from the pool are re-allocated when their sizes are not sufficient for the current call. The (hopefully reasonable) assumption is that RecvMsgs and SendMsgs are called with a somewhat uniform number of Messages/Buffers. If not, the buffers in the pool will likely be re-allocated a few times until they reach the size required for the "biggest" caller. This pooling can significantly reduce the temporary allocations made in recv/sendMsgs. While the effectiveness of this change clearly depends on the usage pattern, pressure on the GC, etc., some improvements can already be observed in the micro-benchmark BenchmarkUDP: the number and size of allocations per iteration decreases significantly and there is a small but noticable improvement in time per iteration. ``` name old time/op new time/op delta UDP/Iter-1-8 5.50µs ± 2% 5.46µs ± 2% ~ (p=0.055 n=20+20) UDP/Batch-1-8 5.65µs ± 3% 5.56µs ± 4% -1.68% (p=0.011 n=20+20) UDP/Iter-2-8 11.0µs ± 3% 11.0µs ± 3% ~ (p=0.645 n=20+20) UDP/Batch-2-8 8.40µs ± 3% 8.24µs ± 5% -1.87% (p=0.012 n=20+20) UDP/Iter-4-8 22.0µs ± 1% 21.9µs ± 3% ~ (p=0.437 n=17+20) UDP/Batch-4-8 13.3µs ± 2% 12.7µs ± 2% -4.69% (p=0.000 n=20+20) UDP/Iter-8-8 44.2µs ± 3% 44.0µs ± 2% ~ (p=0.551 n=20+20) UDP/Batch-8-8 24.2µs ± 4% 23.1µs ± 4% -4.65% (p=0.000 n=20+20) UDP/Iter-16-8 87.9µs ± 4% 88.1µs ± 3% ~ (p=0.708 n=19+20) UDP/Batch-16-8 45.6µs ± 4% 44.1µs ± 5% -3.12% (p=0.000 n=20+20) UDP/Iter-32-8 175µs ± 4% 176µs ± 4% ~ (p=0.087 n=20+20) UDP/Batch-32-8 87.9µs ± 1% 84.5µs ± 6% -3.78% (p=0.000 n=19+19) UDP/Iter-64-8 353µs ± 3% 352µs ± 3% ~ (p=0.414 n=20+20) UDP/Batch-64-8 172µs ± 4% 172µs ±11% ~ (p=0.157 n=20+20) UDP/Iter-128-8 705µs ± 3% 699µs ± 4% ~ (p=0.142 n=20+20) UDP/Batch-128-8 345µs ± 2% 343µs ± 6% ~ (p=0.134 n=20+20) UDP/Iter-256-8 1.41ms ± 3% 1.41ms ± 4% ~ (p=0.758 n=20+20) UDP/Batch-256-8 692µs ± 4% 685µs ± 4% ~ (p=0.114 n=20+20) UDP/Iter-512-8 2.82ms ± 2% 2.81ms ± 2% ~ (p=0.820 n=20+20) UDP/Batch-512-8 1.27ms ± 3% 0.75ms ± 6% -40.62% (p=0.000 n=20+20) name old alloc/op new alloc/op delta UDP/Iter-1-8 408B ± 0% 424B ± 0% +3.92% (p=0.000 n=20+20) UDP/Batch-1-8 440B ± 0% 232B ± 0% -47.27% (p=0.000 n=20+20) UDP/Iter-2-8 816B ± 0% 848B ± 0% +3.92% (p=0.000 n=20+20) UDP/Batch-2-8 696B ± 0% 280B ± 0% -59.77% (p=0.000 n=20+20) UDP/Iter-4-8 1.63kB ± 0% 1.70kB ± 0% +3.92% (p=0.000 n=20+20) UDP/Batch-4-8 1.22kB ± 0% 0.38kB ± 0% -68.42% (p=0.000 n=20+20) UDP/Iter-8-8 3.26kB ± 0% 3.39kB ± 0% +3.92% (p=0.000 n=20+20) UDP/Batch-8-8 2.26kB ± 0% 0.59kB ± 0% -73.76% (p=0.000 n=20+20) UDP/Iter-16-8 6.53kB ± 0% 6.78kB ± 0% +3.92% (p=0.000 n=20+20) UDP/Batch-16-8 4.34kB ± 0% 1.01kB ± 0% -76.75% (p=0.000 n=20+20) UDP/Iter-32-8 13.1kB ± 0% 13.6kB ± 0% +3.92% (p=0.000 n=20+20) UDP/Batch-32-8 8.50kB ± 0% 1.84kB ± 0% -78.34% (p=0.000 n=20+18) UDP/Iter-64-8 26.1kB ± 0% 27.1kB ± 0% +3.92% (p=0.000 n=20+20) UDP/Batch-64-8 16.8kB ± 0% 3.5kB ± 0% -79.16% (p=0.000 n=20+20) UDP/Iter-128-8 52.2kB ± 0% 54.3kB ± 0% +3.92% (p=0.000 n=20+20) UDP/Batch-128-8 33.5kB ± 0% 6.8kB ± 0% -79.56% (p=0.000 n=20+16) UDP/Iter-256-8 104kB ± 0% 109kB ± 0% +3.92% (p=0.000 n=20+17) UDP/Batch-256-8 66.7kB ± 0% 13.5kB ± 0% -79.77% (p=0.000 n=20+20) UDP/Iter-512-8 209kB ± 0% 217kB ± 0% +3.92% (p=0.000 n=20+20) UDP/Batch-512-8 121kB ± 0% 15kB ± 0% -87.89% (p=0.000 n=20+20) name old allocs/op new allocs/op delta UDP/Iter-1-8 14.0 ± 0% 14.0 ± 0% ~ (all equal) UDP/Batch-1-8 14.0 ± 0% 8.0 ± 0% -42.86% (p=0.000 n=20+20) UDP/Iter-2-8 28.0 ± 0% 28.0 ± 0% ~ (all equal) UDP/Batch-2-8 20.0 ± 0% 10.0 ± 0% -50.00% (p=0.000 n=20+20) UDP/Iter-4-8 56.0 ± 0% 56.0 ± 0% ~ (all equal) UDP/Batch-4-8 32.0 ± 0% 14.0 ± 0% -56.25% (p=0.000 n=20+20) UDP/Iter-8-8 112 ± 0% 112 ± 0% ~ (all equal) UDP/Batch-8-8 56.0 ± 0% 22.0 ± 0% -60.71% (p=0.000 n=20+20) UDP/Iter-16-8 224 ± 0% 224 ± 0% ~ (all equal) UDP/Batch-16-8 104 ± 0% 38 ± 0% -63.46% (p=0.000 n=20+20) UDP/Iter-32-8 448 ± 0% 448 ± 0% ~ (all equal) UDP/Batch-32-8 200 ± 0% 70 ± 0% -65.00% (p=0.000 n=20+20) UDP/Iter-64-8 896 ± 0% 896 ± 0% ~ (all equal) UDP/Batch-64-8 392 ± 0% 134 ± 0% -65.82% (p=0.000 n=20+20) UDP/Iter-128-8 1.79k ± 0% 1.79k ± 0% ~ (all equal) UDP/Batch-128-8 776 ± 0% 262 ± 0% -66.24% (p=0.000 n=20+20) UDP/Iter-256-8 3.58k ± 0% 3.58k ± 0% ~ (all equal) UDP/Batch-256-8 1.54k ± 0% 0.52k ± 0% -66.45% (p=0.000 n=20+20) UDP/Iter-512-8 7.17k ± 0% 7.17k ± 0% ~ (all equal) UDP/Batch-512-8 2.61k ± 0% 0.56k ± 0% -78.48% (p=0.000 n=20+20) ``` Fixes golang/go#26838 Change-Id: Id34e362737455cd48df5bc751426be49fbc28094 GitHub-Last-Rev: 4e33d507bb0325fd2cd8b58b586081de1acdf2ad GitHub-Pull-Request: golang/net#102 Reviewed-on: https://go-review.googlesource.com/c/net/+/315589 Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com> TryBot-Result: Go Bot <gobot@golang.org> Trust: Tobias Klauser <tobias.klauser@gmail.com> Reviewed-by: Ian Lance Taylor <iant@golang.org>
This repository holds supplementary Go networking libraries.
The easiest way to install is to run go get -u golang.org/x/net. You can also manually git clone the repository to $GOPATH/src/golang.org/x/net.
This repository uses Gerrit for code changes. To learn how to submit changes to this repository, see https://golang.org/doc/contribute.html. The main issue tracker for the net repository is located at https://github.com/golang/go/issues. Prefix your issue with “x/net:” in the subject line, so it is easy to find.