On Fri, Apr 4, 2008 at 4:03 PM, Carlo v. Loesch <[EMAIL PROTECTED]> wrote:
> Carlo v. Loesch typeth: > | more gain in MUC and pubsub, but you can't say that 50% protocol > > Maths wins over memory: 60% of 70% is 42%, not 50%. > > Since pubsub and MUC are hardly relevant, XMPP has a problem > with 42% of all XMPP stanzas on the network being redundant overhead. > > That's why I repeatedly tend to focus on that problem first. > > +1 Even if you keep all the fancy enhancement proposals outside simple XMPP transfers presence stanzas to remote servers in a redundant form, like Carlo already explained. Google experienced similar problems and they have an immense number of users. Okay, they mostly have to handle with presence stanzas internally but if XMPP gets even more adopted in open communities federation will definitely increase. We need to come up with a multicast like solution to keep traffic and computation time low. Stanza Repeaters are one of many, already proposed, may they've been better or not, solutions and if we don't provide such people why either move to other systems/develop new ones or, if we have luck, develop own multicasting enhancements to XMPP. The first case seems more likely. I also support the idea of a general solution for the multicasting problem rather then developing improvements to all XEPs that may need multicasting. Presence, MUC, PubSub, PEP, and so on only differ in number of subscription changes over time (from low in Presence to high in MUC protocol) and number of stanzas over time. A general solution might not be best at all but fairly good for each protocol which exists now or will exist in future. Tobias
