[Standards-JIG] proto-JEP: Smart Presence Distribution

Carlo v. Loesch CvL at mail.symlynX.com
Wed May 17 17:38:21 UTC 2006


Jean-Louis Seguineau typeth:
| >I thought a quickstart to happiness with a seriously simple JEP
| >already helps us a lot, considering that these presence fan outs
| >are the highest load on the XMPP network from what I saw on some

There once was a statistics page of one popular server, which showed
the popularity of stanza types in comparison. Presence stanzas ranged
at a 3x rate above message stanzas and the rest followed. The page
has been taken down in the meantime. I remember storing a snapshot
of that page, but I packed a couple of archives since then, so
finding it isn't trivial. Can't you just make some new statistics,
so that we can see how big the effect is? We're not the ones running
big Jabber servers. We can only make statistics of how the traffic
looks like _after_ the treatment we are suggesting.

Not that it were really important how big the reduction is, it will
be relevant anyway, and since we are not trying to use a JEP-0033
like approach, using smart unicast or later multicast is a pure win
operation.

Generally, switching to multicast always means losing some abilities
which make information non-multicastable. You will encounter this
problem no matter which multicast strategy you opt for - even if you
sit down for a year and come up with something different to ours.

| >statistics. Obviously excessive users of presence-out wouldn't be
| >exactly useful to the process, but I suppose a majority of users
| >never makes use of that function.
| 
| traffic observations? I personally have a different experience, and the list
| would certainly be interested in understanding better the context in which
| you came to these conclusions. 

Alright, you're the ones running the big Jabber servers. Find out
how many users are actively using presence-out (Simply run a stats
script over your database, right?). I said I was supposing
and that's what it was - a supposition. In the end it doesn't matter,
because we're only talking about trading in traffic for less traffic.
It's always an improvement, even if it only reduced overall traffic by 5%,
although my guess is a lot higher. But you could tell that.




More information about the Standards mailing list