[Standards-JIG] PROPOSAL for Live Chat
dave at cridland.net
Sun Jan 7 11:56:50 UTC 2007
On Sat Jan 6 09:03:23 2007, Ralph Meijer wrote:
> On Sat, Jan 06, 2007 at 12:57:44AM +0100, Yann Le Boulanger wrote:
> > [..]
> > whay I understood is that when I receive a new event, all chars
> > offset are keept, all chars after offset are removed and replaced
> > the one in "partial set of characters" so in you case if I
> receive "6"
> > with offset 3, I keep "123" and replace "456" by "6" so it's
> > that 4 and 5 have been removed
> I'm not sure all of this really is an issue. Remember that the
> ends up in a TCP packet, with a size slightly less than 1500 bytes.
> messages (when finished) are way smaller. So it doesn't really
> matter if
> you send one character or the whole thing.
No, TCP packets vary in size - they typically have a maximal size of
1500, imposed by an MTU/MRU on the IP layer, although fragmentation
and other tricks can change that. But still, a small payload leads to
a small TCP packet. In turn, this leads to penalization in some
networks, leading to higher loss and latency - I think this is
Nagle's algorithm, but perhaps someone else more versed in that layer
might chip in here.
I think if you're aiming at a live chat protocol, you can more or
less drop any pretence of having a network efficient protocol. I
would personally opt for dropping the entire notion, but if not, it
seems reasonable to consider at least moving this out of the band of
XMPP entirely, so at least you're not going to strain S2S links.
Jingle is the obvious option here.
Dave Cridland - mailto:dave at cridland.net - xmpp:dwd at jabber.org
Infotrope Polymer - ACAP, IMAP, ESMTP, and Lemonade
More information about the Standards