MQSeries.net :: View topic

sorn73 · Posted: Tue May 08, 2007 2:58 pm Post subject:

Hello.

I am having a problem with workload balancing in WMQ6.0.2.1 on Windows Server 2003 that I am unable to solve.

I have a cluster, let's call it CLUS1, that contains two Full Repositories, MGR1 and MGR2, and two Partial Repos, MGR3 and MGR4. MGR1 and MGR3 are on the same physical node, the others are on separate physical nodes.

I have created a queue that is shared in the cluster CLUS1 from both MGR1 and MGR2, called TEST. DEFBIND is set to NOTFIXED on both instances. I have not changed any workload balancing parameters.

If I send 10 messages from MGR3 to TEST, they are devided up evenly, 5 messages on TEST@MGR1 and 5 on TEST@MGR2. Like a clockwork, never any problem.

However, if I send 10 messages from MGR4 they are NOT deivded up evenly. I get something like 9 on TEST@MGR1 and 1 on TEST@MGR2. If I then try to send another ten messages, I always get 10 on TEST@MGR1 and 0 on TEST@MGR2. If I then do a Refresh Cluster Information and send another 10 messages, it does a little workload balancing, but then it stops again. I've seen 8/2, 9/1 and 10/0. It never starts to balance again until I do another Refresh Cluster Membership.

It's always MGR1 that is selected, never MGR2. If I do a Suspend from cluster on MGR1 and send another 10 messages, they all end up nicely on TEST@MGR2.

As far as I can see, all channels are up and are showing green arrows everywhere. I have no messages in any event-logs that would point to any channel bouncing up and down, that would explain why they were excluded from the cluster.

So, do you have any hints? Is there any way that I could find out how a workload balancing decision was taken for a given message? Are there any other statuses that I could look at that would give me a hint?