[erlang-questions] How many processes do you run in your BEAM?

Gleb Peregud <>
Fri Feb 14 17:11:16 CET 2014


We did run around 1.7MM processes handling around 850k concurrent TCP
connections. Although it was in a stress test environment. 32 cores of
the physical server were utilized 85% and more under the stress test.
We had spikes in run queue lengths since most of the processes were
"activated" with a single broadcast message. But in our case we were
able to maintain sub 1s latency for all processes to finish processing
of the broadcast message.

On Fri, Feb 14, 2014 at 12:56 AM, Geoff Cant <> wrote:
> I'm curious how many processes people have luck with running in their BEAM vm.
>
> The Logplex cluster at work is running R16B02 with the following stats: (rq = run queue length, procs = number of processes, otp = version number - these figures are reported machine by machine every minute aggregated over 15 minutes)
>
> avg(rq)   avg(procs)   stdev(rq) stdev(procs) max(rq) max(procs) mode(otp)
> 97.792722 50676.159810 74.305638 185.676264 267     50963      R16B02
>
>
> Are there known scaling limits for the schedulers? Are there guidelines for common problems (our avg(rq) becomes greater than some threshold, or as a ratio over avg(procs) it crosses a threshold)?
>
> Curiously,
> --
> Geoff
>
>
>
>
> _______________________________________________
> erlang-questions mailing list
> 
> http://erlang.org/mailman/listinfo/erlang-questions



More information about the erlang-questions mailing list