[JDEV] jabber users limit !!
Vijay Kumar
vijay at dsfinternet.com
Tue Dec 12 06:32:14 CST 2000
well U can sownload the scale.tgz
from http://download.jabber.org/cvs/scale.tgz
it contains the simulator dirctory
as
[root at ws70 simulator]# pwd
/home/vijay/scale/testing/simulator
[root at ws70 simulator]# ls
CVS config.c error.h scalesimulate3.zip simulate3.h
Makefile config.h rand.c simulate.cfg simulator
NOTES error.c rand.h simulate3.c
compile using make it will make the "simulator"
then U can change the settings of "simulate.cfg " according to your need
such as mine setting is :
I have installed simultor on other machine say 192.168.1.170
I was able to get 1024 simulteneous connection with follwing setting .
but I am figring out how to increase it ,through jpolld . presently I am
testing only jabberd
as it is with users system on disk without jpolld .
mine whole idea is to get how much a single machine can take a load
so that when we do server farm then we should have a fair idea of how much
jabber can support . well so far I havent reached any breaktrough to cross
the limit
of 1024 (message_delay=1000 & no message) even with using ulimt (to increase
no of simultenous TCP connection from default of 1024 ).
server=192.168.1.7
server_name=192.168.1.7
port=5222
users=40000
message_delay=1000
login_delay=1
login_duration=300000
user_letter=r
message= hi!!
vijay
----- Original Message -----
From: Ryszard Sommefeldt <ryszards at xnl-interactive.com>
To: <jdev at jabber.org>
Sent: Tuesday, December 12, 2000 2:38 PM
Subject: Re: [JDEV] jabber users limit !!
> Hi, as well as developing a client, my company is also
> looking to run our own servers. So, this debate about the
> maimum user limit is very interesting to us.
>
> We have pretty much 'bet the farm' on Jabber and are
> commited to using it now after considering alternatives.
>
> We have no problems investing in high-end server hardware
> to run our Jabber servers, provided we can reliably sustain
> a large amount of users with 24/7 uptime.
>
> So, would it be possible to post the configuration of the
> system that had 40K users, in as much detail as possible?
>
> Also, what was the testing methodology? How was 40K
> users simulated? Has that ever been reached in a public system?
>
> This would let people interested in the scalability of Jabber,
> have a go at setting their own similar system to see if they
> get similar numbers.
>
> Thanks,
>
> Ryszard
>
> ----- Original Message -----
> From: "elijah wright" <elw at stderr.org>
> To: <jdev at jabber.org>
> Sent: Tuesday, December 12, 2000 6:41 AM
> Subject: Re: [JDEV] jabber users limit !!
>
>
> >
> > > RE: [JDEV] jabber users limit !!Still million $ question is how to
configure
> > > & tune jabber to reach its maximum users limit (40,000) .
> >
> > > users=1000
> > > message_delay=1
> >
> > umm, is that 1000 users sending one message every second?? *chuckle*
> >
> > two things:
> >
> > 1) to assume that every user will send a new message every second is
nuts-
> > most people will have large blocks of idle in there. i type very
> > quickly; i probably send an IM message at a clip of one every three
> > seconds at peak rate. usually, though, i'm not donig that. i'm usually
> > idle.
> >
> > so as a basic metric, multiply your 1000 tested live users by three.
then
> > probably multiply that by 15 or so due to the amount of lag time in most
> > conversations. i'd think differently if it was groupchat, but simple
IM's
> > shouldn't be putting such a huge load on there.
> >
> > 2) 1000 database-backed operations in a second is absurdly fast. temas
is
> > a magician or a djinn or something to get a PIII/700 to be so pimp ;)
> > i'm thinking especially about the kind of performance you get out of an
> > apache server with a mysql backend; its not that fast. :)
> >
> >
> > if you really need better performance than you're getting (and how do
you
> > know without testing it?) you should probably be looking at bigger
> > hardware :) :) :) *loves the hardware freakshow*
> >
> > my guess would be that 40,000 users each sending one message every
> > second would probably eat several bonded 100BaseT connections to a
*very*
> > large server or server farm... if that's what you really need you
> > probably need a bigger box than anything wintel. (i'm a solaris bigot,
> > personally; pick your scalable unix of choice)
> >
> >
> >
> > elijah
> >
> >
> > _______________________________________________
> > jdev mailing list
> > jdev at jabber.org
> > http://mailman.jabber.org/listinfo/jdev
> >
>
>
> _______________________________________________
> jdev mailing list
> jdev at jabber.org
> http://mailman.jabber.org/listinfo/jdev
More information about the JDev
mailing list