[Tsung] Suggested method of testing a Jabber Servers max user connects

[Jason Tucker]

Oh, and some other not-so-obvious things to keep an eye on when trying to
run a high number of concurrent connections:

The Tsung machines need to have enough ephemeral TCP ports available to
establish a high number of concurrent connections. Not sure what OS you
have, or how many Tsung hosts your using, but you might need to tweak this
in your system if you start running out of ports. This link might be useful
in that regard:

http://www.ncftp.com/ncftpd/doc/misc/ephemeral_ports.html

Similarly, whatever XMPP server your testing may need to be tweaked to
provide enough filehandles to support all the simultaneous connections.

__Jason

On Feb 4, 2008 10:53 AM, Jason Tucker <jasonwtucker at gmail.com> wrote:

> Session duration is definitely a factor. If your sessions are configured
> to last for say, 60 minutes (using various thinktime periods dispersed
> around your session), and your connect rate is 10/sec, then at the end of
> the first 60 minutes of the test, you should have ~36,000 connected users.
>
> Not sure which Erlang your using, but if it is one of the single-threaded
> versions, then you have to be really careful to watch your CPU utilization
> on the Tsung controller. Even on a multi-cpu box, the single-threaded erlang
> will only use 1 CPU or core for the controller, and once that controller
> process maxes out that CPU, you're done. Doesn't matter how many Tsung
> client processes your are using to spread the load, the controller process
> will be a bottleneck in that situation. I've gotten around this problem in
> the past by simultaneously running multiple Tsung instances (with separate
> sets of users for each controller instance).
>
> SMP erlang should help solve that controller bottleneck problem, but I
> haven't don't have too much first-hand experience with that yet.
>
> __Jason
>
>
> On Feb 4, 2008 8:28 AM, Taylor Laramie <taylor.laramie at globalrelay.net>
> wrote:
>
> > I'm attempting to use Tsung to bench/stress a replacement system for our
> > current Jabber server (ejabberd) to get an idea what of the maximum
> > number of concurrent user connections it can handle w/ light load. The
> > main consideration is the use of TLS/SSL which I've been able to get
> > Tsung to use but now I'm getting an fatal error at around 9000 users
> > before it completes the tasks.
> > Does anyone have any suggestions or an example tsung.xml for testing
> > load capacity for an xmpp server (using TLS) ? The main items I'm
> > testing for are :
> > - max con. users
> > - rate of connection
> > - generous amount of activity (message sending, roster interaction ,etc)
> > To wit I'm using interarrival to control rate and global_number +
> > userid_max maximum concurrent.
> >
> > Regards,
> > --
> > Taylor Laramie
> > Systems Administrator
> > _______________________________________________
> > Tsung-users mailing list
> > Tsung-users at lists.process-one.net
> > https://lists.process-one.net/mailman/listinfo/tsung-users
> >
>
>
>
> --
> Please help support my participation in Philly's Livestrong Challenge:
> http://philly08.livestrong.org/tuckerjw
>
> Find me on Facebook:
> http://www.facebook.com/profile.php?id=1016862963
>



-- 
Please help support my participation in Philly's Livestrong Challenge:
http://philly08.livestrong.org/tuckerjw

Find me on Facebook:
http://www.facebook.com/profile.php?id=1016862963
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.process-one.net/pipermail/tsung-users/attachments/20080204/10f43a84/attachment.html