Clustering and total ordering

[timfox]

I've been thinking some more about how we're going to implement clustering and how this relates to total ordering protocols.

It seems to me, that to successfully implement clustering for JMS we need to replicate our state machines: channel (queue/subscription) state, transaction state etc., and this requires a total ordering protocol.

It's my understanding that JGroups currently implements a total ordering using a single node (synchronizer), and a token based total ordering is in the pipeline.

Bela - am I on the right track? I'd be interested to get your input here.

[belaban]

Yes. We have SEQUENCER (which is preconfigured in seq.xml), which I wrote recently. And we have TOTAL_TOKEN which is token-based and which Vladimir wrote some time ago (coincidentally he joined JBoss 2 weeks ago...).

If you are looking to use seq.xml, I'll schedule it for some serious performance and regression testing. Let me know so I can plan ahead (JIRA-wise)

[timfox]

Thx.

I need to read up some more on the pros & cons, differences between all these.

How does TOTAL_TOKEN relate (if at all) to http://jira.jboss.com/jira/browse/JGRP-6? Is this a different total token protocol?

[belaban]

JGRP-6 is a re-implementation of TOTAL_ORDER. Let's have a call on FRI if you're available (I'm onsite right now)

[timfox]

K. Let's do that