Re: ejabberd vs. Tigase

Thank you both for your responses.
I managed to run a few test myself using Tsung, simply looking at how well
each handled rapid connections to the XMPP server (20K connections).
Unfortunately, I couldn't get Tsung clients to receive messages from a
specific client (seems Tsung only supports bidirectional messages for
subscription requests only at the moment). From just managing connections,
it seems Tigase has a much higher CPU usage, ~30% CPU for an inter-arrival
time of 0.005s up to 20K connections, compared to ejabberd's ~3% CPU usage
for the same test. Memory usage was not nearly as different, but ejabberd
(500 Mb) still used less than Tigase (750 Mb). Comparing against Artur's
500K test, my numbers seem reasonable (he was getting ~11% CPU, but with a
much beefier server). Do any of these number seem inaccurate to anyone?
(Hardware is a two quad-core sun fire x4150, I don't have the more detailed
specs off-hand)

We likely won't be using PubSub, since at broadcast time, the recipient list
will be dynamic. Is PubSub much more efficient than sending a message one by
one to each client? I heard both ejabberd and Tigase supporting millions in
pubsub projects, but often over a number of nodes. We will only have a
single node, potentially a second node at max. Anyone have deployment
numbers/limits on only one or two nodes?

Thanks,

Dan

On Mon, May 25, 2009 at 3:33 AM, Mickael Remond <
[email protected]> wrote:

> Hello,
>
> Tomasz Sterna wrote:
>
> > Dnia 2009-05-06, śro o godzinie19:47 -0400, Dan Le pisze:
> >> Has anyone worked with both ejabberd and Tigase and could offer
> >> insights on how they compare in terms of performance, resources used
> >> (CPU, RAM), community support, ease of use, and the like?
>
> I can only speak for ejabberd, but there should be more than 10000
> deployments around the world running ejabberd and ranging from very
> small to very large deployments.
>
> Regarding consumption, ejabberd is probably the one that can be tuned
> to the smallest size. We have been running ejabberd with 500
> simultaneous users on a Linux PDA.
>
> Regarding the large size deployments, we have many very large
> deployments (currently working on several one supporting more that a
> million simultaneous online users). You can tune it and tweak it to fit
> your need. You can do thing which are not possible with other server in
> a very short amount of time.
>
> Regarding pubsub as well, feedback from current deployment is very
> good. Supporting millions of nodes with large number of items is
> possible. It is also very flexible thanks to its plugin architecture.
> We plan to give more details soon on the very large scale pubsub
> projects we are working on.
>
> You can contact me privately if you want to discuss your project in more
> details.
>
> --
> Mickaël Rémond
>  http://www.process-one.net/
> _______________________________________________
> JAdmin mailing list
> FAQ: http://www.jabber.org/discussion-lists/jadmin-faq
> Forum: http://www.jabberforum.org/forumdisplay.php?f=19
> Info: http://mail.jabber.org/mailman/listinfo/jadmin
> Unsubscribe: [email protected]
> _______________________________________________
>

_______________________________________________
JAdmin mailing list
FAQ: http://www.jabber.org/discussion-lists/jadmin-faq
Forum: http://www.jabberforum.org/forumdisplay.php?f=19
Info: http://mail.jabber.org/mailman/listinfo/jadmin
Unsubscribe: [email protected]
_______________________________________________