[Foundation] Jabber Klatch: Discuss .org server role

Max Metral Max.Metral at peoplepcHQ.com
Thu Aug 9 11:52:40 CDT 2001


That could be a great answer.  Somebody is reporting dropped messages, from
this message it would seem that we WOULD spend time fixing that if we can
right?

-----Original Message-----
From: Thomas Muldowney [mailto:temas at box5.net]
Sent: Thursday, August 09, 2001 12:48 PM
To: members at jabber.org
Subject: Re: [Foundation] Jabber Klatch: Discuss .org server role


Well to be fair to myself and the work I've done you can press the open
source
server pretty high if you use dpsm or jpolld from cvs.  Then there is
sheath's
farming work, which I'm sure allows it to go beyond that.  So the groundwork
for high user counts is there (scaling is another issue).  

http://www.box5.net/~temas/10k.jpg

There is an image of the open source server running with 10k users
connected,
granted the mem usage is pretty high, but this was back in the 1.2 range,
and
before release optimizations.  Haven't run recent tests, but Dustin has been
(granted he's had probs, but it's not an easy task to even get some boxes 
themselves to accept 10k connections).

So the server itself can last for a little while longer.  While it is
maintained
primary development is focussing more on the Jabelin (http://www.jabber.org)
setup.  Don't jump ship and think that this is competition in any way,
rather
it is a development team (with some fresh ideas) for the next gen jabber.org
system.  Our immediate design goals are to be scalable (although not 
neccessarily having high user counts immediately), seperate out functional 
pieces, and rethink as much as possible.  Because development focus is
shifting
many people are seeing a stall, but it's not there.  Granted there are only
a 
few people working on jabelin, but jabberd was mostly written by jer and a
few
touches here and there by other people.  Because it's always been a small
group
working on the server jabelin wants to work in teams on different aspects of

the system (XDB, config manager, SCM, CCM, HUB, etc) and utilize the entire
group as a design and philosophy base.

To answer the primary/original question, jabberd will probably not scale
much 
more than it already can (10k roughly?), but the jabelin system should be
able
to scale as much as you are willing to throw hardware at it (farming, and
all).

Thoughts, questions, ideas?

--temas

On Thu, Aug 09, 2001 at 08:12:12AM -0700, Max Metral wrote:
> No, not all do.  AOL, MSN for example (not that they matter in this case).
> PeoplePC, large but not at that scale, doesn't either.  Nobody wants to
have
> max at colorado.peoplepc.com...
> 
> -----Original Message-----
> From: johnston at megaepic.com [mailto:johnston at megaepic.com]
> Sent: Thursday, August 09, 2001 11:06 AM
> To: members at jabber.org
> Subject: Re: [Foundation] Jabber Klatch: Discuss .org server role
> 
> 
> Well, large ISPs usually break things down into cities and stuff, no? I
> mean, @home sure does. Could we design a gateway server that keeps records
> of what user is on what server and forwards appropriatly?
> 
> Mat.
> 
> On Thu, Aug 09, 2001 at 07:26:20AM -0700, Max Metral wrote:
> > Your example works for internal use maybe, but what about an ISP with a
> > couple hundred thousand users, or EBay or any of those sorts?  Right now
> > .org is not an option for that.
> > 
> > -----Original Message-----
> > From: johnston at megaepic.com [mailto:johnston at megaepic.com]
> > Sent: Thursday, August 09, 2001 1:09 AM
> > To: members at jabber.org
> > Subject: Re: [Foundation] Jabber Klatch: Discuss .org server role
> > 
> > 
> > Well, large organizations tend to be just that, organizations...
generally
> > there have already been divisions made up between departments and
offices,
> > etc. In this respect, the jabber system scales pretty well (even though
> one
> > particular server may not) since you can have user at dept.company.com or
> > something along those lines. It would of course be good to improve the
> > server whenever possible, but a huge number of users do not need to be
> > supported as an inital design goal; rather, a more appropriate design
goal
> > would be to minimize the effort required to improve upward scalability
> > should it be required. What kinds of things would stop the jabber server
> > from scaling to large user counts? If these things are identified, why
can
> > they not be fixed? Are there large compromises between management effort
> and
> > scalability?
> > 
> > Mat.
> _______________________________________________
> Members mailing list
> Members at jabber.org
> http://mailman.jabber.org/listinfo/members
> _______________________________________________
> Members mailing list
> Members at jabber.org
> http://mailman.jabber.org/listinfo/members



More information about the Members mailing list