[Standards-JIG] Multicast JEPs in the making

Carlo v. Loesch CvL at mail.symlynX.com
Wed May 17 17:12:16 UTC 2006


Jean-Louis Seguineau typeth:
| You just have to convince the list that your findings are bringing all the
| advantages you mentioned in your nice message. We are all ears. 

It's only a month ago or so that a discussion on this list came to figure
out that a common multicast scheme for presence, MUC and pubsub would
solve a lot of scalability issues. We are not here to repeat what has
already been digested.

| I myself would be particularly interested by some metrics about the
| scalability you have achieved with your prototype. 

PSYC has been using this strategy for presence for over a year and
the multicast bits have been in use (for high traffic chat) ever
since 2000. The adaption to XMPP protocol was trivial and we simply
saw that it works the same way.

PSYC is in use for the yearly European Music Awards of MTV and has
been in use for high load VIP chats with T-Online for several years.
We had guests like Harrison Ford or Robbie Williams day-in day-out.
Scalability was never an issue with our technology, so we stopped
measuring. In particular simulated load checks weren't satisfactory,
because the mere load is no problem - the problem can be the latency
and instability effects of the network, and it is quite a piece of
work to emulate that. But we still have the test scripts lying around
somewhere, should this really be considered useful information. So
far PSYC was always capable to cope with any client's needs, no matter
how large the audience he would bring.
 
| P.S. What is "your IETF experience" btw?

I took part in a couple of IETF meetings around 1997 when the IMPP-WG
was created. When the IMPP-WG didn't understand how much IM is about
one-to-many and how useless it is to leave the 'chat' aspect out of
the design, I abandoned the working group. The same reasoning also
kept me out of Jabber until now.




More information about the Standards mailing list