Gabble vs Salut (Re: What's left for Update.1)

Morgan Collett morgan.collett at collabora.co.uk
Sat Feb 2 08:31:25 EST 2008


Jim Gettys wrote:

> I think we need to seriously discuss about possibly/probably being
> update.1 fodder is the "kids arrive at school in the morning" problem.
> 
> 5 - Use of mesh in large, crowded environments
> If everyone arrives at school running local link and resumed quickly,
> the network might melt from mdns mesh traffic's interaction with the
> mesh's implementation of mutlicast.  We've upped the multicast bitrate
> for multicast as a band aid, until we can dynamically adjust the
> bitrate.  But the fundamental issue comes that in large, dense school
> environments, can't expect multicast to scale far enough, and should be
> using unicast to a presence server (jabber in our current case) to
> handle this problem.
> 
> Dave Woodhouse has suggested may be to try to get a response to the
> school server's anycast address, and if we get a response from a school
> server, switch from Salut to Gabble for presence service automatically.

Rob logged #6299 about this (presence service should disable salut in
the presence of school servers on mesh). We should try the schoolserver
before enabling Salut.

He proposed an algorithm and some open questions regarding timeouts. We
need some feedback on those, and more info on the anycast address.

> Another temporary solution would be to have Ohm ask NM to reconnect if
> the machine is suspended for more than some interval, say, 30 minutes.

Possibly related: I logged #6304 regarding my MP XO not being able to
ping (or anything) after a long sleep (on AP, gabble). Ohm nudging NM on
wakeup would probably fix this too.

Morgan



More information about the Devel mailing list