On Mon, 11 Dec 2017 22:08:03 +0000, Farley, Peter x23353 wrote: > >I do not see any "modulus" operation available in the SORT manual, so it >apparently isn't possible to post-process the IEBDG numbers with SORT to >convert each of the random numbers modulo the limit value. I suppose division >and truncation and subtraction may work, though I haven't figured out how yet. > Modulus has been considered an inferior way to generate random numbers within a given range because with many RNGs the low-order bits are less random than the high-order. For example for some congruential RNGs the low order n bits simply repeat with a cycle of 2**n. Better to multiply and take the integer part.
Modern RNGs with cryptographic strength may not suffer this deficiency. The better suggestion was to generate random keys and sort. >The other hard problem is to guarantee that there is exactly one occurrence of >each number and that no numbers in the range are left out. > Longer random keys greatly reduce the probability of duplicates. How does performance of DFSORT vary with key length? (TANSTAAFL) -- gil ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
