Performance Gain For arsload

[Ed_Arnold]

Some arsload jobs have found using HEAPPOOLs rather than default storage usage helpful in reducing elapsed time and/or CPU consumption.

You set this for an individual arsload job by using the ENVAR LE runtime option, which can be set via a CEEOPTS DD statement.                   
                                                                                                                                                                                                       
 //CEEOPTS  DD *                                                           
 ENVAR('ARSMVS_ARSADMIN_LEOPTS=RPTSTG(ON),HEAPP(ON)')                                       
                                                                         
This will cause a usage of HEAPPOOLs, which is a different heap management algorithm for smaller sized heap elements.

The only known downside is it *might* use a little more storage.

As always, no guarantees, test first, etc etc etc, but especially if you have a long running arsload it's worth a try.

This is good on both 7.1.2.x and 8.4.x.x.

I would appreciate it if anybody tries this that they give me some feedback.

Ed

[bormson]

Hi Ed. I did some testing with this using a fairly large AFP report with 62,000 documents. I did not see any difference in performance. Can you verify that it is HEAPP(ON rather than HEAP(ON)?

[Ed_Arnold]

Hi Benny - I went back to the original problem to see what type of data the load was performing but that information is no longer available.

Dump analysis from while the load was running showed frequent calls to LE for an internal working storage routine.

Changing the working storage routine to use HEAPPOOLS resulted in the improvement.

The original problem was that moving from 7.1.2.5 -> 7.1.2.11 resulted in a many times increase in CPU consumption.  Adding the HEAPPOOLS parm put CPU consumption back to the expected value.

Ed   

[Ed_Arnold]

This is very short, step by step instructions on Tuning heap storage

http://publibz.boulder.ibm.com/cgi-bin/bookmgr_OS390/BOOKS/CEEA2170/3.3.3.4?SHELF=CEE2BK71&DT=20060529020347

If anybody out there can find the time to try this out, I'm sure I and many others would love to hear your results.

Ed (trying to find a round tuit myself to try some of these things that I learn about)