And don't benchmark from windows either, at least on Windows 10 we've 
measured the latency at ~200ms, turning off nagle, caching, everything does 
not effect it.  Bench from a linux VM on the same box and that 200ms 
vanishes and we get 2ms latency over the internal network again...


On Saturday, July 23, 2016 at 7:16:20 AM UTC-6, Scott Ribe wrote:
>
> On Jun 10, 2016, at 12:29 AM, Matt Hornsby <[email protected] 
> <javascript:>> wrote: 
> > 
> > But when i hit it with wrk, with 100 open connections from the mac, i 
> get: 
>
> 1) Phoenix cannot measure the time that it takes a client to establish a 
> connection, while of course any benchmark from outside will be accounting 
> for that. However I would expect the discrepancy to be more like 2ms. 
>
> 2) wrk can be slower than Elixir/Phoenix ;-) (I suspect that it's got some 
> poor code around the threading, but anyway when I benchmarked with it I 
> established that latencies were much more strongly correlated to number of 
> threads per wrk node than they were with number of threads to the phoenix 
> node--of course that requires more than 2 machines to test, and is a real 
> pain. But I would say that I don't know why people like wrk so much, I 
> found it to be very weak. Anyway, I can assure you that with 100 threads 
> you absolutely ARE experiencing some delay in wrk itself.) 
>
> 3) When you use exrm, don't forget MIX_ENV when building the release, 
> otherwise you just bundle the dev build, if you've previously compiled for 
> prod. 
>
> -- 
> Scott Ribe 
> [email protected] <javascript:> 
> http://www.elevated-dev.com/ 
> https://www.linkedin.com/in/scottribe/ 
> (303) 722-0567 voice 
>
>
>
>
>
>

-- 
You received this message because you are subscribed to the Google Groups 
"elixir-lang-talk" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/elixir-lang-talk/3e3a7080-dfc3-49dc-9b87-2046edeaf851%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to