[erlang-questions] Ssl tuning

[Dan Gudmundsson]

We saw that on the client side, (and came up with same things as you)
I havn't seen the same issue on the server side, but maybe I just
managed to load the server hard enough.

Give me a couple of days and I'll get back to you

/Dan

On Fri, Apr 13, 2012 at 5:24 PM, Sergei Fomin <dafgar@REDACTED> wrote:
> Hello.
>
> We tried some load testing of erlang ssl and have some questions.
>
> The bench is Xeon with 24 cores, otp R15B and modified mochweb server.
> Mochiweb acceptor is modified to do gen_tcp:accept, start new acceptor and
> call ssl:ssl_accept. Original mochiweb acceptor calls ssl:transport_accept,
> ssl:ssl_accept and only then spawns new acceptor.
> The server accepts requests and answers 200 immediately.
>
> First iteration.
> When load rate reached 1300 rps ssl_connection_sup process began to
> accumulate message queue.
>
> Second iteration.
> I moved all ssl_connection initialization code from init to separate
> function, the call to which was triggered by message send from init (simply
> delayed init). This way ssl_connection_sup was just spawning ssl_connection
> process and sending init message to it.
> Ssl_manager process became the new bottleneck. The threshold load rate
> increased insignificantly.
>
> According to top all cores had load ~60%.
>
> Any ideas how to tune ssl noninvasively to achieve more results?
>
> --
> Sergei Fomin
>
> _______________________________________________
> erlang-questions mailing list
> erlang-questions@REDACTED
> http://erlang.org/mailman/listinfo/erlang-questions