MQSeries.net :: View topic - Upgrading to Tandem G06.25 and Up !!!! WARNING !!!!

hankknowles · Posted: Tue Nov 15, 2005 4:33 am Post subject:

For HP Nonstop systems utilizing MQSeries V5.1 an issue that did have a severe impact under heavy message load.

We went to G06.25 and the Monday following this change a quirk began to appear. It manifested itself in the following way

1. Large quantities of messages were hanging in the queues but QPASA failed to identify a problem as the messages were indeed flowing but ever so slowly.

2.Process $MQ01 ( mqs-queue00) server's primary process was litterally in a long wait stat using minimal cpu or pages. While the backup process was claiming huge quantities of memory pages. The utilization on the backup process started running into the 80- 90% range.

3.The cpu that was running the backup process finally isssued a software halt took it's dump and reloaded.

4. Once we put humpty dumpty together again we discovered that the heap size comparison between primary and backup were vastly different the backup process' heap size was 8 times the size of the primary.

5. The software halt was caused my TMF process and it hung in a weird state and there was a transaction that would not end.

Upon investigation by HP and IBM support we found that the heap size was out of range between the two process', which is the culprit probably caused by the new native heap manager on the G06.25 O/S 1269 which you have to decide to use in the first place and we chose to go that route as we are running Java proc's to do encrypt and decrypt of all data in and out of this application. These process' really tax our system.

As it turns out if you are planning on going above G06.25 and you are using the new heap manager 1269, you need to make sure that you have CSD03 as well C3EFIX3 installed. Once these are in place you should be good to go!

Good Luck and be careful out there!