DFSORT called by DB2 utility (DSNUTILB) abending ICE083A

dbecker · Posted: Tue Feb 24, 2009 8:23 pm

Hello, I keep getting the following error within DFSORT...

dbzTHEdinosauer · Posted: Tue Feb 24, 2009 8:58 pm

I would start by trying to solve:
ICE083A D RESOURCES WERE UNAVAILABLE FOR DYNAMIC ALLOCATION OF WORK DATA SETS (970C)

dbecker · Posted: Tue Feb 24, 2009 9:03 pm

dbzTHEdinosauer · Posted: Tue Feb 24, 2009 9:54 pm

This is the document you need for an explanation of ICI083A

SC26-7525-02, Z/OS DFSORT Messages, Codes and Diagnosis Guide.

you can also find this by using the IBM LOOKAT site at url:
www-03.ibm.com/servers/eserver/zseries/zos/bkserv/lookat/

dbecker · Posted: Tue Feb 24, 2009 10:04 pm

Dave Betten · New User Joined: 24 Jan 2006 Posts: 26

First let's look at your space requirements. With an lrecl of 50 and FILSZ=E000001384787124 you have total file size of about 64GB. We usually estimate the work space requirement as about 1.3 times the file size so you're looking at around 80GB. That equates to almost 100,000 cylinders. So based on how much free space you have on individual volumes, you can determine how many work data sets you need. If you have a lot of volumes with at least 3000 cyls available, you can probably get by with 32 work data sets.

Now how do we get DFSORT to use 32 work data sets? For a DB2 Utility, you cannot override DYNALLOC via DFSPARM. You need to pass SORTNUM 32 in the utility control statements and this will cause the utility to pass DYNALLOC=(,32) to DFSORT. This is documented in info APAR II14047 which provides a whole bunch of useful information on using DFSORT with DB2 Utilities.

Another thing I would mention about this utility is that I see MAINSIZE=65536K is being passed. This sort would probably benefit from more virtual storage because it's so large. DB2 uses DFSORT's installation default for DSA as the maximum mainsize it will pass. So if you have a number of DB2 utilities with sorts of this magnitude, you should consider having your DFSORT installation default changed from its shipped value of DSA=64 to DSA=128.

dbecker · Posted: Wed Feb 25, 2009 12:16 am

Thanks Dave!

I have 16 volumes in the pool that is being used for the sortwk datasets. This consist of 6 mod27's with 32000 cyl each free and 10 mod9's with around 9000 cyl each free. So should I specify SORTNUM 32 in the utility control statements. I cannot go over 32 as this causes dfsort to not use the blockset technique...correct? With plenty of space in this pool but only 16 volumes....is this an issue for dfsort? There are 5 indexes on this table so that means 5 sorts....does dfsort then take the sortnum times the number of indexes? I really appreciate the help?

Dave Betten · New User Joined: 24 Jan 2006 Posts: 26

I would first try SORTNUM 12. DB2 is going to pass DYNALLOC=(,12) to each of the sorts. If DB2 is able to maximize it's parallelism, I expect we'll have 6 concurrent sorts (1 for the data, 5 for the indexes). So that means we'll spread 72 work data sets across 16 volumes. You could try a smaller sortnum but that means fewer/larger work data sets. I think you have a better chance of them all fitting if we allocate more/smaller data sets. I don't see a way to avoid multiple work data sets on the same volume given the size of this utility and the number of volumes available.

Also, fyi, DFSORT blockset can support up to 255 work data sets. There is no limit of 32 for blockset.

dbecker · Posted: Wed Feb 25, 2009 2:52 am

Dave,

I tried the following JCL as to allocate my own sortwk...

dbecker · Posted: Wed Feb 25, 2009 3:03 am

In the new job that I'm adding the SORTNUM 12 on the db2 control statement I will be removing all the allocated sortwk dd's in my JCL and let DFSORT do whatever it needs....Just wanted to make sure everyong is on the same page.

Dave Betten · New User Joined: 24 Jan 2006 Posts: 26

Can you download the entire sysout (not just the dfsort messages) to a text file and email it to me? My email is betten@us.ibm.com

Dave Betten · New User Joined: 24 Jan 2006 Posts: 26

Earlier we estimated that this sort would need about 100,000 cylinders of work space. You coded 10 SORTWKnn statements all with a primary and secondary allocation of 4350 cylinders. So we started out with 43,500 cylinders of primary space. We received B37 abends on all 10 when trying to obtain additional secondary space. I just realized that's because you can't use more than 64K tracks (about 4369 cylinders) on any of these data sets unless you add DSNTYPE=LARGE to the SORTWKnn dd statements. This will cause them to be allocated as large format and thus be able to use additional space on the MOD9 and MOD27 volumes. DFSORT supports large format and in fact DFSORT's dynamic allocation allocates them that way. But when you use JCL work data sets, you have to specify DSNTYPE=LARGE to get them allocated as large format.

How's the run with SORTNUM 12 going?

dbecker · Posted: Wed Feb 25, 2009 4:10 am

OK..good info to know. The SORTNUM 12 using DFSORT's dynamic allocation is running now. It will take it some time so I'll post the results later tonight or in the morning. Thanks.

dbecker · Posted: Wed Feb 25, 2009 6:08 pm

Dave,

Sent you a couple emails with the sysouts of the last 2 runs. I'm currently running with my own work dsn allocations and have included the DSNTYPE=LARGE on each of those dd's. If you see anything that raises concern on those sysouts please let me know....I'm willing to try about anything at this point.

Thanks.

dbecker · Posted: Wed Feb 25, 2009 6:31 pm

Dave,

Got your reply on the SORTNUM 12 still not passing DYNALLOC to DFSORT. Will wait to hear more. Thanks.

Dave Betten · New User Joined: 24 Jan 2006 Posts: 26

Got a reply back from my colleague in DB2 development. According to the DB2 Utility Guide "If you omit SORTDEVT, SORTNUM is ignored" Sorry I was not aware of this. You need to specify

SORTDEVT SYSDA SORTNUM 12

This should cause the sort to use 12 work data sets instead of the default of only 4.