[ejabberd] some ejabberd memory benchmarks
christopher.zorn at gmail.com
Mon Aug 18 18:39:54 MSD 2008
On Mon, Aug 18, 2008 at 9:36 AM, Fabio Forno <fabio.forno at gmail.com> wrote:
> I'm doing some benchmarks of jabber servers and I've been astonished
> about memory occupation of ejabberd, while all other parameters
> (latency and throuput) are really fine.
> My setup is ejabberd 2.0.1 with mysql backend under linux. I've
> created two sets of users:
> 1) 30k users with 1-10 (random) contacts each one (10r)
> 2) 30k users with 1-150 (random) contacts each one (150r)
> Roster contacts are spread in a widow of 1000 and 2000 consecutive
> users, respectively, so that if I go online with a set of consecutive
> users must of each user's roster becomes online.
> The testing clients are made with twisted, but I don't think this
> influences the test.
> Here are the results with 10k users online:
> * 10r, login only: 614MB
> * 10r, login and go online: 962 MB
> * 150r, login only: 1148MB
> * 150r, login and go online: 1940 MB
We are noticing similar things, are these tls connections? I ask because
searching for this I found the following ticket.
> 75 contacts per user in the roster are not a lot, and 1.9 GB of memory
> for 10k online users seem too much...
> Am I missing some tuning parameters or these are normal figures for
> ejabberd installations?
> Fabio Forno, Ph.D.
> Bluendo srl http://www.bluendo.com
> jabber id: ff at jabber.bluendo.com
> ejabberd mailing list
> ejabberd at jabber.ru
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the ejabberd