we have an issue with one of our Cobol programs that calls DFSORT, I'm not a cobol programmer so I don't really understand what the issue is....the guy whos investigating it says :-
basically they have made the main stoarage handling super ace including 64K , but our programs are address mode 24 , so because its forced to use the main storage, when there is more than 16mb of data in its internal sort file the job abend S0C3 cause it cant address over 16mb with AMODE 24
The programmer is asking if there is any way to change TUNE= default parameter ?
IIRC, TUNE parameter cannot be overridden via DFSPARM.
If a change is indeed required to be made to this, then it has to be done through ICEPRMxx members in the site PARMLIB.
A more detailed discussion is provided in the Customizing DFSORT section of the 'DFSORT Installation and Customization' guide.
But to summarize, it CAN NOT be overridden from JCL; and what you're looking for probably is TUNE=OLD; a site admin can help you identifying the site PARMLIB (from past experience some sites customize this).
I'd leave it to DFSORT experts to answer this query in detail as they'd have more experience with this.
Though, from past experience what you can do is:
a. Restrict memory usage by putting in control over MOSIZE/HIPRMAX; this can be managed through DFSPARM
b. Add a decent sized DYNALLOC value - word of caution --> before you override this, would recommend that you do have someone look at the defaults, AND visit the DFSORT manual to understand how the changed values will behave; DFSORT guide has a good chapter on calculation of work-space
c. Simply, add the SORTWK DD statements to the JCL; if they are present, IIRC, they are allocated to DFSORT (by JCL rules they are allocated during run-time in any case; but DFSORT refers them as well; usage to a higher degree is governed by the restriction on the MOSIZE workspace, which is requested to be limited as per point-a)
Lastly, please do perform regression testing with a fairly large dataset (read At Least a few million records and a large RECL), and various combinations of MOSIZE/HIPRMAX and work dataset allocation; because it does tend to vary site-to-site.