[standards-jig] [JDEV] Re: Publish-Subscribe JEP0060

Timothy Carpenter timbeau_hk at yahoo.co.uk
Mon Dec 16 08:38:51 UTC 2002


On PS JEP-0060

Just a first scan comment.

1) Permissioning may need some alternatives - though I can see this model
being useful in some areas.

a) The publisher may wish to contribute to who can/cannot see the data but
not be the owner - e.g. in a scenario of who can see data could be "anyone
but Microsoft". This can be done with 'in band' information within the
payload but something says that might be storing up trouble for the future.
Although the publisher could be given Owner permissions, this is too coarse
and you may not want the publisher to be able to do all the activities of an
owner.

b) The JEP keeps permissions separate (i.e. this is handled off-camera by
the owner) but as the concept is mentioned, it should be seen just as a
possible mechanism. No realtime system would survive if a server had to go
back to the initial owner for each new subscription! (see further points
below)

2) There is no provision for stale data alerts (warning subscribers that the
publisher is no more), heartbeats or gap filling.

3) A publisher coming 'online' could be asked to provide the full image for
all current persistent nodes so the server can provide the service.

Many of the above questions are pretty much follow-ons from my own work in
pub-sub. 

The big issue to me is scalability and republishing. I am very much a
believer of a 'fractal' approach to this problem - i.e. that a server can
appear as a publisher to another server and that a server can appear as a
user to another server. In such a situation, the 'owner' and 'publisher'
changes depending on how deep you go in the hierarchy. For scalability it is
important for the publisher, server and owner to delegate. Without it things
can get messy.

As the owner can change we need to be able to classify data inband to state
what it is so that it can be permissioned. As each level gets the data, it
may want to use other means of classification or convert classifications to
other mechanisms (e.g. when collating data from different publishers using
different permissioning systems).

A diagram is often the best way to convey this concept, but the point is
that permissioning is tied in with pub-sub and, apart from some more
straightforward situations (rooms), having an owner specify JIDs or just use
a JID as the means for permissions with no other information at hand simply
will not fly. We need JIDs to be given some form of authorization that
interlocks with permissioning coding of the individual data items or groups
thereof.

Permissioning and pub-sub are therefore interlinked and thus will need to be
considered sympathetically.

brgds
Tim

__________________________________________________
Do You Yahoo!?
Everything you'll ever need on one web page
from News and Sport to Email and Music Charts
http://uk.my.yahoo.com




More information about the Standards mailing list