[ejabberd] Ejabberd hardware requirements for very high load

Michael Weibel michael.weibel+xmpp at gmail.com
Wed Jan 11 15:50:32 MSK 2012


Hi David & Kirk,

>
> I'm actually presuming that they are still using ejabberd, yes partly
> because of the erlang bit in their articles, but also because they were
> previously looking for ejabberd experienced developers and I THINK they
> might have mentioned it a few years ago (when they started).
>
> It would be nice to hear a bit more detail about their setup, although I
> suspect that might count as company secrets :)
>

To hear about their setup would be absolutely nice and I hope someone of
them is reading this. :)


> Bear in mind that from the articles they are running those 2 million
> connections on a single machine.
>
> ..which is pretty impressive!


> Unfortunately one of the best guides to setting up ejabberd clustering,
> seems to have gone offline
> http://liferayz.com/how-to-setup-clustering-in-ejabberd-tutorial/ perhaps
> someone has an archived copy of that ? (its not on the wayback machine I
> checked).
>

I just setup 2 machines with ejabberd & clustered together. It's sad that
the documentation about it is pretty bad.. In the ejabberd guide it sounds
pretty easy but actually only half of the process is described there.
Using this link:
http://tdewolf.blogspot.com/2009/07/clustering-ejabberd-nodes-using-mnesia.html
And this conversation:
http://chatlogs.jabber.ru/ejabberd@conference.jabber.ru/2006/08/04.html

I was able to get the clustering working. It would be pretty nice if there
would be some more detailed documentation on the ejabberd website.

Regarding my original question: It's a little bit hard to guess what
whatsapp actually did to get to those 2 million connections, although they
mentioned their machine (which has a lot of cpu's and high memory). I think
they did additional tuning to reach that goal.

I'd be glad if anyone else could tell their experience regarding this.
But maybe I need to setup some small clustering system with a loadbalancer
and do a benchmark on top of it...
Does anyone know if this guide: http://www.ejabberd.im/benchmark is still
the way to go for benchmarking?

- Michael
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.jabber.ru/pipermail/ejabberd/attachments/20120111/a5233cb3/attachment.html>


More information about the ejabberd mailing list