Re: [hybi] WS port with BOSH-like fallback?

[Jack Moffitt <[e-mail]>]

> Has anyone tried measuring the connection success rate - and
> messages/packets/data/latencies - of "valid" HTTP port 80 techniques:
> such as single-connection bidirectional chunked POST, and various
> dual-connection pipelined-POST/GET (one upstream, one downstream)?

I don't have any scientific measurements, but I did run a chess
community of about 400k users, and we experienced close to 100%
connection success rate as far as I could tell.  This was using XMPP
BOSH, which on a client typically uses the same two sockets for the
life of the page, and exchanges full HTTP requests and responses
containing XML.  Compression is even used almost always from server to
client, but only on certain browsers in the opposite direction.

My anecdotal experience was that except in the very rare cases were
intermediares have forced timeouts on connection length, everything
worked without a hitch (modulo temporary network failure).  In the
cases where connections were forced closed after N seconds (where N <
60 or so), things still worked, but latency was terrible.

Our success rate for direct XMPP connections with the desktop client
was not as good, since many places block everything but port 80 and
443.  It was always our intention to have the desktop client fall back
to BOSH since BOSH seemed the most reliable.

jack.