[Standards-JIG] Re: pubsub meeting 2004-06-23

Bob Wyman bob at wyman.us
Wed Jun 23 23:05:33 UTC 2004


I think the list should be extended to include a need to discuss potential
mechanisms for optimizing s2s transmission of messages to multiple
receivers. This was mentioned during the meeting and has been a subject of
discussion between Peter Millard and me in the past.

The problem is this:
	The server may discover that it needs to send the same message to
more than one person on another server. What it wants to do is identify not
only the recipients of a message but which subscriptions have been matched.
The subscriptions identifiers will be different depending on the recipients
and individual recipients may need to be notified of more than one matching
subscription.
	What the server would like to say is:
	Deliver a copy to tom at example.org for his "sub1" subscription.
	Deliver a copy to peter at example.org for his "sub3" and "sub4"
subscriptions.
	Deliver a copy to mary at example.org for her "332k2" subscription.

	Delivering to multiple JID's should be done using JEP-0033, however,
JEP-0033 doesn't provide a means to transport the subscription ids. The
subIDs can be identified using SHIM, but SHIM can't be used for addressing.
	What may be required is to define an inter-server s2s redistribution
or proxy protocol that provides a means of combining addressing with
extended headers.
	Providing this capability would be very helpful in reducing overall
network bandwidth utilization as well as in reducing the delivery load on
servers in the chain from pubsub service to client. Such a "fan-out"
capability would be able to use hierarchies or meshes of servers very
efficiently in delivering messages. If we look forward to a time when we may
have 100's of thousands of people subscribing to pubsub services, it should
be clear that eventually, we'll be able to save massive amounts of work for
all parties if we can distribute the load of delivery using something like
what is sketched above.

		bob wyman




More information about the Standards mailing list