[ejabberd] clustering and memory usage
jesse.thompson at doit.wisc.edu
Wed Oct 24 17:40:35 MSD 2007
> 2007/10/22, Jesse Thompson <jesse.thompson at doit.wisc.edu>:
>> I've been playing around with "-env ERL_FULLSWEEP_AFTER 0", which seems
>> to help keep the process from growing too much. Are there any downsides
>> to this?
> There are several paragraphs in this page which seem interesting, look
> for 'fullsweep':
> In summary it may save RAM and consume more CPU.
Interesting. I'll play around with different values and do a more
thorough analysis of the effects.
>> Would it help to change some or all of the "disc_copies" mnesia tables
>> to "disc_only_copies"? I think that this might help if the roster table
>> is especially large.
> At least 'offline_msg' is not required in RAM I guess.
OK, well, I guess I'll experiment.
It would be helpful if there was a guide explaining more about each
mnesia table, explaining what it's used for. This would help make the
process of deciding whether to copy it across cluster nodes and whether
it should be kept in RAM, just one disc, or both.
>> Any other ideas?
> You can try to use pgsql or mysql as storage instead of mnesia.
I thought I read that there were similar memory issues even if an
external database was used.
> And now speaking in the long term. There is an proposal to modify data
> types used internally in ejabberd: 'use binaries instead of lists
> where possible':
> I made such a change in some of my Erlang code non-ejabberd-related,
> and the program consumes less RAM and a lot less CPU. In the case of
> ejabberd, I don't know how difficult and beneficial it will be, but it
> will be interesting to see.
> ejabberd mailing list
> ejabberd at jabber.ru
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 3340 bytes
Desc: S/MIME Cryptographic Signature
Url : http://lists.jabber.ru/pipermail/ejabberd/attachments/20071024/c63d3893/attachment.bin
More information about the ejabberd