Slow Spec Merge on Unix,Xe

dhelsley

Well Known Member
HpUx, Xe, sp14.2, Oracle 8.1.5

We have a situation that causes our queues to backup. I was wondering if anyone else has seen it. Or, if anyone with a similar environment could test it for me I'd appreciate it.

Some of our vresions of R43500 have a large number of sections with ER and layout overrides. Because of this the specs are sent to our enterprise server (.pak) and merged into the enterprise server specifications before running the UBE. While this is happening an exclusive lock (a mutex in SAW) is issued for the UBE and version. This lock prevents another iteration of the same UBE and version from attempting to merge their specs at the same time.

Now here's the problem. Merging the specs for this UBE takes, on average, 90 seconds (I've seen it take 3 minutes). During our peak we have dozens of users submitting this job to our multithreaded queue at about 1 every 15 seconds. Each of these locks the next out for 90 seconds. Soon we have every thread of our queue filled with the same UBE/version in an "S" status. The mutex only blocks the same UBE and version, but since all of the threads are waiting to run this report nothing else can get in.

To create the situation just override the ER for every section in a version of R43500 and submit it with logs. You can tell if you have the same situation by checking the time spent "Starting" your UBE. This is the time between Startup and RUNBATCH. Notice the time between those two in this bit of jde.log. The actual job ran in 8 seconds. Merging the specs took three minutes.

14702 Tue Jun 5 10:42:41 2001 runbatch.c618
Startup for User=DHELSLE, Env=DV7333, Job#=73481

14702 Tue Jun 5 10:45:42 2001 runbatch.c1066
RUNBATCH: Remote CP=1252, Remote OS=5, Local CP=1252,

14702 Tue Jun 5 10:45:50 2001 ipcpub.c3214
API ipcSawUnregisterProcV1 : process 14702 unregistered inIt

This occurs on all three of our Unix enterprise servers. We've tested it on an NT app server and the delay is 2-3 seconds instead of 90.

I use R43500 as an example here. I've tested other UBEs with substantial amounts of ER logic. The result is the same.

Thanks for your help.

David D. Helsley, Inc.
Independent IT Consultant
[email protected]
<P ID="edit"><FONT SIZE=-1>Edited by dhelsley on 6/8/01 04:36 AM.</FONT></P>
 
Re: Slow Spec Merge on Unix,Xe *DELETED*

Post deleted by dhelsley
 
David,

have you thought about applying the changes to the base template instead of the version (since you've customized everything anyway)?

This may eliminate your problem. Our R43500 is also highly customized but we applied the changes to the template and are experiencing no problems.

I wasn't aware that version overrides forced recurring spec merges on a server. This seems strange. I assume you've deployed the version(s) in a server update package?

Regards,

Larry Jones
[email protected]
OneWorld B733.1, SP 11.3
HPUX 11, Oracle SE 8.1.6
SandBox: OneWorld XE SP15
 
Yes, full packages have been deployed to the enterprise server and all TSEs. I even tried setting InstallSpecs=N on the client. This seems to have no affect.

You can see the .pak files under .../spec/<username> in the default path. Of course, if things are working, you can only see them before they are unpacked and deleted.

I have a number of alternatives in mind including changing the template. This client has a "zero mod" mind set so changing the template is something I'd need to convince them to do. They have a small mod to the P4210 that they've reapplied 14 times so far. Gotta love ESUs. Xe makes some of this easier.

Thanks for the input.

David D. Helsley, Inc.
Independent IT Consultant
[email protected]
 
Back
Top