Cluster design doc updates.

[Alan Conway <ac...@redhat.com>]

Updated cluster design doc
http://cwiki.apache.org/confluence/display/qpid/Cluster+Design+Note

This design is aimed strictly at reliability/HA. It's a lot simpler and
more implementable than previous designs and addresses thorny issues in
more detail (error handling, the dequeue problem) that were glossed over
before. I think it works active-active or active-passive as a
matter of configuration, performance experiments will show where the
advantage lies.

The things I've taken out vs. earlier designs (fragmented queues,
proxies etc.) can be seen as a layer on top of reliability. Any scheme
that can federate or cluster individual brokers for throughput or
distribution reasons can equally be applied to a set of reliable broker
clusters.

Any thoughts much appreciated.

Cheers,
Alan.