In a message dated 8/26/2005 1:29:15 P.M. Central Daylight Time, [EMAIL PROTECTED] writes:
>What number might you specify to optimize elapsed and/or CPU time without "being ridiculous"? Barry Merrill presented a superb paper at SHARE in Anaheim in FEB 84 on this subject. If you can find that paper in the SHARE archives, you would do well to read it. First thing you must do is define "optimize", meaning optimize what metric at the possible expense of all other metrics? You mentioned two metrics - (1) elapsed time and (2) CPU time. There are many complicating factors. You may be able to "optimize" your own performance but everyone else running in the same z(ed) image suffers terrible performance while your application runs. E.g., if you code BUFNO=huge, for very large values of huge (>>5), then you will tie up large amounts of real storage which, in some situations, may cause other users to suffer excessive page faults. Depending on your installation's charge-back scheme, you may be severely penalized for the extra large REGION requirement you would then have as well as for a very large amount of (real storage)*(occupancy time). You will also possibly hog the channels, thus causing delays to other users. In general, you can optimize with respect to one variable at a time, but not very often more than one. Another way to say this is that if you push the system in at one point it may well bulge out at another point. Read Barry's superb 21-year-old paper for starters. There are many, many variables to consider, many of which are spawned by the Law of Unintended Consequences. Bill Fairchild ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
