Performance metrics?

[richardanaya]

Would be nice to publish some metrics

[jpsamaroo]

While a benchmark of Mux would really just be a benchmark of HTTP.jl, if you'd like to benchmark Mux and post the results here, we could certainly find a way to publish it :)

[jpsamaroo]

I take back what I just said about Mux having the same performance as HTTP.jl, I no longer believe that would be the case under load. @richardanaya would you be able to perform some initial benchmarking and submit a PR with the benchmark script and results?

[richardanaya]

Of course, I'll get some this week

…

On Sun, Apr 21, 2019 at 12:38 PM Julian Samaroo ***@***.***> wrote: I take back what I just said about Mux having the same performance as HTTP.jl, I no longer believe that would be the case under load. @richardanaya <https://github.com/richardanaya> would you be able to perform some initial benchmarking and submit a PR with the benchmark script and results? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#92 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AACHZGSAPY4S2QEAOTP4XM3PRS7DRANCNFSM4HHA5QHQ> .

[richardanaya]

richardanaya@penguin:~/dist$ ab -n 100000 -c 100 http://100.115.92.203:8080/
50% 36
66% 43
75% 53
80% 60
90% 96
95% 223
98% 248
99% 255
100% 469 (longest request)

richardanaya@penguin:~/dist$ ab -n 10000 -c 1000 http://100.115.92.203:8080/
50% 44
66% 48
75% 51
80% 56
90% 162
95% 3379
98% 4015
99% 4409
100% 4870 (longest request)

richardanaya@penguin:~/dist$ ab -n 10000 -c 10000 http://100.115.92.203:8080/
50% 2352
66% 2357
75% 2365
80% 2371
90% 2375
95% 2682
98% 4377
99% 4578
100% 7956 (longest request)

I noticed failures for high n and high c counts

[jpsamaroo]

Can you elaborate on what these numbers are, and provide instructions for how one could run these benchmarks on their own machine (including the Julia code to setup the Mux server, and whatever packages need to be installed to get this ab command)?

Thanks for following up on this! Once we have all the details we can turn this into a PR.

[richardanaya]

My performance approach might be wrong, after reading this article, I want to try a different method

https://github.com/aj-monk/C10k.jl