[ejabberd] Clustered eJabberds not load-balancing

Oliver Schmidt oliver.schmidt.wue at arcor.de
Tue Dec 22 17:14:14 MSK 2009


Greetings,

I've found the reason: I had SMP on both nodes enabled which resulted in  
massive context switching (more than 2000 switches/s) and slowed down the  
cluster. Now, I've disabled SMP and the cluster is 6x faster but the  
second CPU core remains unused. Hmm, too bad.

Oliver

Am 22.12.2009, 09:27 Uhr, schrieb Oliver Schmidt  
<oliver.schmidt.wue at arcor.de>:

> Greetings,
>
> I have two cluster environments, each consisting of two ejabberd nodes.  
> When I watch one cluster's behaviour, it seems that it distributes CPU  
> load between each node equally. The other cluster, however, seems only  
> to share the Mnesia database but each node seems to handle its own load  
> (seemingly depending on the users connected to that node).
>
> Do you have a clue what might be wrong?
>
> Thanks in advance and merry christmas
>
> Oliver


More information about the ejabberd mailing list