AW: AW: [ejabberd] ejabberd vs Jabber XCP
ulrich.staudinger at lycos-europe.com
Fri May 5 16:12:01 MSD 2006
In my tests I came to a different memory consumption.
That's the memory chart, ramp up time of 20 minutes, staying there for another 20 minutes.
Left is memory in MB, right number of conc. Users.
Von: ejabberd-bounces at jabber.ru [mailto:ejabberd-bounces at jabber.ru] Im Auftrag von Joel Reymont
Gesendet: Freitag, 5. Mai 2006 13:34
An: ejabberd at jabber.ru
Betreff: Re: AW: [ejabberd] ejabberd vs Jabber XCP
On May 5, 2006, at 12:21 PM, Pieter Rautenbach wrote:
> It's indeed a 64-bit server. I understand the reasoning, but I
> disagree: not all
> ejabberd data in memory could be strings! Thus, the memory usage
> cannot be entirely attributed to the use of strings. Here are similar
> figures for a single CPU server, having 14,065 users online:
> USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> ejabberd 16 0 4778m 3.7g 1692 R 23.3 95.8 63:00.41 beam
I never meant to imply that all the memory is taken by user contexts, i.e. data kept per user. All I'm saying is that ejabberd, Mnesia, etc. use strings internally and most everywhere it appears.
You should still see 1-2Gb of memory use on a 32-bit machine and a 64- bit machine probably does not gain you much right now. As soon as string use within ejabberd is optimized (Summer of Code project?) you should be able to fit twice as many users on your 64-bit server. I don't think a factor of 16 (1 byte vs 16 right now) is likely, though.
ejabberd mailing list
ejabberd at jabber.ru
More information about the ejabberd