Page 1 of 1

IPC timeout issue

Posted: Thu Jun 21, 2012 1:52 pm
by mdbatra
Hi All-
I understand from a search here on the same topic that it has been discussed umpteen no. of times, but still I am not able to come to a conclusvie statement regarding the root cause and the resolution.

We refreshed out test enviornment(SMP,8 node) with the production one(SMP, 4 node). There are so many jobs making use of IPC stage which run fine in production(with 512 buffer size and timeout as 10 ) but all such jobs are failing in test environment one after another. We tried with changing the timeout to 60 and that has worked fine.

Not able to understand:
1. how do the jobs run with similar configuration in production ?
2. Is there a way to propogate the buffer size and timeout as set in administrator to the IPC stages? As there are lot of jobs and it would required a continuos monitoring of the load otherwise.
3. What's the link b/w setting at project, jobs and IPC stage level ? we have timeout in admin set as 600, at job level we have 'Use Project Defaults' but with in IPC stage we have as 10.
4. What i understood from documentation is IPC(inter-process) works for MPP, while in-process buffering for SMP. so as we have SMP here, shouldn't in-process buffering been used while designing jobs ?

DS v 7.5.3

Posted: Thu Jun 21, 2012 4:41 pm
by ray.wurlod
First question is about resources. Are the production and non-production environments identical (CPU power, memory, disk space)?

Buffer sizes propagate when objects are newly created but not otherwise. There's no link - last one configured wins.

Posted: Fri Jun 22, 2012 3:05 pm
by mdbatra
Hello Ray-

I came across one of the IBM resolution(setting OLD TIMEOUT BEHAVIOR =1) at project level. Post addition of this variable, the jobs are running smoothly now.

I have requested the infrastructure architect to provide the details for the production and test unix boxes. Will get back soon with facts.

Posted: Fri Jun 22, 2012 3:59 pm
by chulett
Interesting... no other post out here (and there are quite a few) mentions "OLD TIMEOUT BEHAVIOR". Where did you find this, can you post a link?

Posted: Fri Jun 22, 2012 4:25 pm
by mdbatra
Sorry Craig - not sure if you meant to be more specific or really asking for the link :)

Nevertheless, this is the link:

http://www-01.ibm.com/support/docview.w ... wg21450066
which i found in the below post:

viewtopic.php?t=145952&highlight=mutex+timeout

Cheers

Posted: Fri Jun 22, 2012 5:43 pm
by chulett
Just really asking for the link. :wink:

Thanks.