Scaling with the current Gradient model
Problem: There's an inherent scaling problem in routing lots of Gradient requests through one TCP stream, which is what happens when you treat a Jabber client as a server. The real Jabber server becomes a bottleneck, albeit a very useful one :-).
-
Possible solution: bust the spec and make multiple requests.
- down:
- obviously braindead.
-
Possible solution: host apps directly on a jabber server, so instead of this:
gradient://user@subdomain.domain.tld/path?fooyou have this:
gradient://subdomain.domain.tld/path?foo
From which the JID "subdomain.domain.tld/gradient" is derived as per URL Semantics.
- up:
- looks familiar...
- fixes problem
- down:
- I need to look @ & fiddle w/ code on jabber servers.
Questions that I will have to look into:
- Is there an identical problem between federated servers?
- Can a Jabber server federate with other jabber servers, therefore providing apps but no auth etc.? i.e. only on 5269?
- Can an XMPP server communicate with another XMPP server using multiple streams over 5269?
© 2006. Some rights reserved. Author: Ian Sollars.
