[ejabberd] ejabberd mod_muc bottleneck

Eric Cestari eric at ohmforce.com
Tue Jun 22 23:46:23 MSD 2010


Le 22 juin 2010 à 21:34, Karthik Kailash a écrit :

> Hi Eric,
>  
> Thanks for your insight.  I have attached my Tsung configuration file.
>  
> So do you think that optimizing mod_muc to use multiple worker processes to handle messages is valid?

Having a single router is a good practice and it can should not be a bottleneck. However spawning processes whenever there's synchronous ops is a good thing. It's already the case for discovery, and should maybe generalized to room creation. More load testing is required :)

> Regarding mochiglobal, would the usage of that be to read in all the mnesia values into memory at the beginning, and then access them from there?
Actually completely bypass mnesia. The code would load ejabberd.cfg directly in mochiglobal.
That should improve performance in many key areas of ejabberd (such as ACLs).

Eric

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.jabber.ru/pipermail/ejabberd/attachments/20100622/6aaa1d9a/attachment-0001.html>


More information about the ejabberd mailing list