>>- I am a little worried about your remark about the (short) life of the >>address spaces: creating a new thread in Unix is just simply adding an >>entry to a process table; if it requires address space creation in z/OS, >>this is a heavy overhead part added to kicking off the task. Can you >>play around with BPXPRM parameters to keep these address spaces (BPXAS I >>suppose) alive for a longer period, so they can be reused? >An idea to keep in mind. I am not really all *that* fluent in unix/bpx >stuff, I didn't even know that that possibility exists. >All our USS workloads tend to write job end iefactrt messages and iefuji >job begin messages into syslog at a very high rate, indicating short >lived address spaces, and obscuring the really important information.:-)
PMFJI, but the last few remarks about being "ported from open world" made me think. So here are a few thoughts related to z/OS UNIX: - z/OS UNIX process are run in UNIX initiator address spaces, aka BPXAS. WLM starts one if the kernel needs while creating a new process. Once a process ends, the BPXAS is going to sleep for 30 minutes, then terminates. (No, you can't change that interval) So, address space creation is only needed if there is no idle BPXAS waiting for work. - In a busy system with quite some UNIX work going on, you'd have a good chance to find an idle BPXAS. However, when the UNIX workload goes down for more than 30 minutes, all those idle BPXAS will terminate. A sudden spike in UNIX work will then lead to a bunch of new BPXAS being created, very expensive, indeed. - Do you see a lot of "BPXAS starting" syslog entries when your workload arrives? If so, this might be part of the problem. - z/OS UNIX offers the possibility of so called "local processes". If your workload is calling spawn() to create the new processed, you might try to set _BPX_SHAREAS=YES and see if this helps. - You might want to try to implement a simple program that does a couple of simultaneous forks() to create a bunch of child processes. They don't have to do anything meaningfull. They can return as soon as all childs have been created. This would ask for so many BPXAS as there are childs. Then when your workload spikes up, the BPXAS are ready. - Finally regarding the iefactrt and other syslog messages: Remember a UNIX process is kind of like a job: SMF records will be cut whenever the process does an exec() and finally when the process ends. SMF exits are driven unless you suppress them for SUBSYS OMVS im SMFPRM. So, seeing all those messages does not mean the address spaces the processes ran in have terminated (see BPXAS above). As long as there are no "BPXAS has ended" messages no UNIX initiator AS has terminated. Hope that helps somewhat -- Peter Hunkeler Credit Suisse ---------------------------------------------------------------------- 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
