[Standards-JIG] proto-JEP: Smart Presence Distribution
Carlo v. Loesch
CvL at mail.symlynX.com
Thu May 25 14:40:20 UTC 2006
Richard Dobson said:
| I see... you want to start this new for every connection, yes it could
| work that way the way you propose it, it does mean that you are only
| likely to get savings where there are lots of contacts between two
| servers that have enough traffic travelling between them to keep the s2s
| connections active for long periods, if they keep regualarly dropping,
| because of various things like lack of activity for a certain period,
| you loose the benefit of this, this is a start but IMO it would be
there is also a special case allowing for connection-timeout events, so
only erroneously lost connections or data losses trigger a retransmission.
| better if we are going to standardise a mechanism for this that it works
| across different connections.
philipp thought it should be a stream feature so we have to wait one
round-trip less and we can start using it rightaway, but your point makes
sense - after all, when evolving this into multicast, it does indeed go
through multiple connections. also, during the first presence and probe
fan-out, it doesn't make much of a difference if the same tcp packet
also contains a disco handshake. a good implementation would send all
three stanzas wrapped into a single tcp packet, compressed or not.
so yes, i am favorable to using disco instead of stream:features.
More information about the Standards