[ejabberd] Identifying pubsub performance bottlenecks

Matthias Rieber ml-erlang at zu-con.org
Mon Oct 15 11:17:53 MSK 2012


Hi,

Am 12.10.2012 13:13, schrieb Evgeniy Khramtsov:
> On 12.10.2012 19:31, Matthias Rieber wrote:
>> Hi,
>>
>> Am 12.10.2012 03:12, schrieb Evgeniy Khramtsov:
>>> On 12.10.2012 07:18, Matthias Rieber wrote:
>>>> ... how can I find the bottleneck.
>>>
>>> p1_prof is your friend here.
>>> Recompile ejabberd with debugtools=true and type p1_prof:help() 
>>> from
>>> the remsh (`ejabberdctl debug`).
>>> You likely need p1_prof:reds(), p1_prof:q() and p1_prof:fprof_* 
>>> functions.
>>
>> thanks! I'm not sure how to interpret the results, may be mnesia's 
>> write speed is the limit? These outputs seem to be interesting:
>>
>> <snip>
>
> Mnesia is definitely the problem in your case: there are some slow
> transactions.
> I'm not an expert in the pubsub code, so Chris could tell more.

EJAB-1461 sounds a bit like the issues I have. Maybe 5000 users with 50 
subscriptions can be considered as "large number of subscribed nodes".

Are there any news about the ejabberd3 release? 
http://blog.process-one.net/road-ejabberd-3/ mention October 2012.

Matthias



More information about the ejabberd mailing list