On Wed, 24 Feb 2021 11:55:18 -0600, Joel C. Ewing wrote: > ... >use a SHR option that allows multiple concurrent opens for write and >constantly refreshes buffers for each record processed. ... > >If the object is to be able to write randomly interspersed records from >multiple jobs running concurrently, another possibility that might be >explored is to use a UNIX file and have all the jobs append output to >the same unix "log" file. > Open the file with O_APPEND (C, REXX SYSCALL, JCL) or ">>" (Shell). I have experienced poor performance appending to a UNIX file. IBM support tells me that is a consequence of need to preserve integrity.
>z/OS assumes you want file integrity and deterministic behavior and >enforces that on MVS data sets. >The Unix world assumes you want what you "ask" for and if you ask for >multiple processes to write to the same file concurrently in ways that >are non-deterministic, it assumes random ordering from concurrent writes >is acceptable. > The randomness will be less for a single write() operation with a a datum < kernel buffer size (131072?) -- gil ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
