When Bill says... > You are correct. I was assuming that nobody had (and > the original poster had NOT) done anything really > inappropriate. For any CISIZE past 16KB you get into > diminishing returns on a 3390, capacity-wise. There > is almost never any good reason to specify a CISIZE > for a LINEAR data set other than 4 KB, especially > not when used how they were originally intended to > be used: with DIV. Assuming a random access pattern, > they work much better, quite a bit more efficiently, > meaning less I/O and substantially less CPU, with a > 4 KB CISIZE (at least according to my own artificial > benchmarks - YMMV).
Y'really ought to take him at his word. These "artifical benchmarks" he has created to test a certain LDS/DIV based component have, over the last few years, beaten the living daylights out of DIV processing to the extent that he can tell you exactly where DIV is spending its time. He even has heuristic algorithms to optimize DIV save processing. It's unlikely anyone outside of system test in Pokey has anything close the insight Bill has into how that puppy works. CC ---------------------------------------------------------------------- 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
