[JDEV] Distributed design of jabber
temas
temas at box5.net
Wed May 2 19:35:38 CDT 2001
Ok, well...
In short: dpsm is the devpoll socket manager. It handles sockets in
the same manner as c2s currently, but does it in a higher scaling
fashion.
In long, normal c2s only handles 1024 sockets, that sucks. So I looked
at the latest and greatest socket technology and saw devpoll to probably
be the best option. I then designed dpsm around devpoll and around the
idea of running it in front of the main jabberd to handle a lot of
sockets.
Example: two mid size boxes in front of a larger backend box, the
smaller running the jpollds, and the larger only jabbed.
So there is a lame summary of it and an example setup. It's still
growing, but should be functional.
--temas
On 02 May 2001 15:06:46 -0500, Colin Madere wrote:
> temas,
>
> Any chance you could give us a quick run-down of what dpsm does and how it
> should be used? I scanned the source and such, but it would be nice to get
> a clear picture from the author about it :)
>
> colin
>
> > -----Original Message-----
> > From: temas [mailto:temas at box5.net]
> > Sent: Wednesday, May 02, 2001 1:22 PM
> > To: jdev at jabber.org
> > Subject: RE: [JDEV] Distributed design of jabber
> >
> >
> > dpsm is me. It works right now (requires kernel module for devpoll
> > support). It's not fully tested, but that should be
> > happening more this
> > week. Give it a go and help develop it and I'll love you =)
> >
> > Not sure on mod_farm right now, haven't looked in a while.
> >
> > --temas
> >
> > On 02 May 2001 09:56:50 -0500, Colin Madere wrote:
> > > Another note:
> > >
> > > According to jabber.com, they charge based on concurrent
> > users on the
> > > server. Currently they are in a state which it is not
> > profitable for them
> > > to deal with small companies(they didn't really give me a number) or
> > > companies that did not have a _gaurenteed_ high number of
> > concurrent users.
> > >
> > > Going on three weeks now that I'm waiting for answers to
> > technical questions
> > > to see if the jabber.com server is a viable solution for my company.
> > >
> > > IOW, until they get some big sales and can beef up on
> > resources (read: more
> > > employees), I wouldn't waste your time or theirs unless you
> > have a need for
> > > a server (and can afford it - once again, I didn't get any
> > prices) that will
> > > have a constant concurrent user count above 10K (that's my
> > guess, since they
> > > didn't give me any solid numbers).
> > >
> > > (Don't get me wrong, I understand their need to focus on
> > customers that can
> > > bring in the big cash until they have good income and are
> > stabilized)
> > >
> > > WHILE I'm on the subject, hey JER and whoever else has
> > worked on it, what's
> > > the status of dpsm and.... 'mod_farm' is it?
> > >
> > > > -----Original Message-----
> > > > From: dizzyd at dizzyd.com [mailto:dizzyd at dizzyd.com]
> > > > Sent: Wednesday, April 25, 2001 8:24 PM
> > > > To: jdev at jabber.org
> > > > Subject: Re: [JDEV] Distributed design of jabber
> > > >
> > > >
> > > > Lemme just translate this marketing stuff to engineering speak:
> > > >
> > > > > >"Enables Distributed Processing across multiple server
> > > > "farms" which in
> > > > > > turn may support multiple CPUs."
> > > >
> > > > Jabber.com reworked the server to be pre-emptively multi-threaded
> > > > (pthreads). Additionally, some work on JSM was done to
> > permit multiple
> > > > JSMs to be fully meshed across a network.
> > > >
> > > > > >"Allows components of a Jabber server to be distributed
> > > > across multiple
> > > > > > machines, enabling a greater degree of inherent redundancy."
> > > >
> > > > Not sure what this means. :)
> > > >
> > > > > >"Enables groups of "socket" connections to be
> > distributed across
> > > > > >mini-servers
> > > >
> > > > The component formerly known as jpoll. Open source now has an
> > > > equivalent called dspm (or dpsm, never can get it straight).
> > > >
> > > > As it stands, j.com has pretty much rewritten most of jabberd
> > > > to be super
> > > > efficient and thread-safe. This was _not_ a minor
> > undertaking, but was
> > > > well worth the effort. Jabber.com provides a super-fast,
> > fully QA'd
> > > > and peer-reviewed implementation of jabberd/jsm/etc. They've
> > > > worked very
> > > > hard to make it scalable and robust. :) Doing these sorts
> > of things
> > > > (scalability and robustness) have not been things that
> > the Open Source
> > > > movement has shown much interest in this far (with good
> > reason). Very
> > > > few people need to run a Jabber server for 200k+ concurrent users;
> > > > these are (some of) the people that Jabber.com caters to.
> > > >
> > > > Hope that helps. :)
> > > >
> > > > Diz
> > > >
> > > > _______________________________________________
> > > > 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