search

jonhoo / volley

Volley is a benchmarking tool for measuring the performance of server networking stacks.
MIT License
123 stars 12 forks source link

Investigate impact of increasing GOMAXPROCS on go server performance #2

Closed jonhoo closed 9 years ago

jonhoo commented 9 years ago

In #1, @tsenart pointed out that increasing GOMAXPROCS might improve the performance of the Go server significantly. This issue tracks that claim.

I'm currently running experiments with GOMAXPROCS = 4x#CPUs to see whether this makes a difference or not.

tsenart commented 9 years ago

Nice, thanks! Also, it'd be good to use the latest stable Go version: 1.4.2

jonhoo commented 9 years ago

Results:

impact of increased GOMAXPROCS

Seems like increasing GOMAXPROCS is only detrimental to the performance of the Go implementation unfortunately.

In terms of the Go version, upgrading is a bit of a pain, and as far as I can tell, nothing performance-related changed in 1.4.1 -> 1.4.2, so it shouldn't matter.

tsenart commented 9 years ago

Thanks for running this experiment. Would you consider sharing these results with the Go team? On Mon 15 Jun 2015 at 22:34 Jon Gjengset notifications@github.com wrote:

Results:

[image: impact of increased GOMAXPROCS] https://camo.githubusercontent.com/0c3a9ef4fde3349f8c56172ebaddc1617dc93e6f/68747470733a2f2f6a6f6e2e746865737175617265706c616e65742e636f6d2f73686172652f766f6c6c65792d676f2d6869676865722d70726f63732e706e67

Seems like increasing GOMAXPROCS is only detrimental to the performance of the Go implementation unfortunately.

In terms of the Go version, upgrading is a bit of a pain, and as far as I can tell, nothing performance-related changed https://github.com/golang/go/issues?q=milestone%3AGo1.4.2 in 1.4.1 -> 1.4.2, so it shouldn't matter.

— Reply to this email directly or view it on GitHub https://github.com/jonhoo/volley/issues/2#issuecomment-112200745.

tsenart commented 9 years ago

By these, I mean the comparison between Go, Rust and C.

On Mon 15 Jun 2015 at 22:37 Tomás Senart tsenart@gmail.com wrote:

Thanks for running this experiment. Would you consider sharing these results with the Go team? On Mon 15 Jun 2015 at 22:34 Jon Gjengset notifications@github.com wrote:

Results:

[image: impact of increased GOMAXPROCS] https://camo.githubusercontent.com/0c3a9ef4fde3349f8c56172ebaddc1617dc93e6f/68747470733a2f2f6a6f6e2e746865737175617265706c616e65742e636f6d2f73686172652f766f6c6c65792d676f2d6869676865722d70726f63732e706e67

Seems like increasing GOMAXPROCS is only detrimental to the performance of the Go implementation unfortunately.

In terms of the Go version, upgrading is a bit of a pain, and as far as I can tell, nothing performance-related changed https://github.com/golang/go/issues?q=milestone%3AGo1.4.2 in 1.4.1 -> 1.4.2, so it shouldn't matter.

— Reply to this email directly or view it on GitHub https://github.com/jonhoo/volley/issues/2#issuecomment-112200745.

jonhoo commented 9 years ago

Yes, I've thought about doing that. I'd like to first run the experiment against go tip though, just to make sure that it hasn't already been fixed. Tracking in #3.