On Wed, 14 Apr 2010 17:39:53 -0500, Arthur Gutowski <aguto...@ford.com> wrote:
>In keeping with the consolidated response (if I'm not losing too much context): > >Mark, > >Yes, I recall the striping issue you had. I think we can avoid that. >Yes, the allocate/delete job is what I referred to. Glad to hear that's not >needed any longer. One less hurdle for our Storage team. Band-Aids can be >tough to remove. In which direction did you adjust the migration threshold? > First, lets rewind a little. This thread was originally discussing the issues related to LOGR in a sysplex with heterogeneous LPARs. Our sharing of LOGR volumes across SMSplexes was a requirement (although the title of this thread is non-SMS LOGR). Making our SVC dump pool shared was done at my request. The development sysplex in this environment is pretty much all static data (SAP DB2 in one LPAR, WAS in the other) and I couldn't get the storage team to implement HSM in that sysplex to manage the dumps. So I pushed them to create the single pool across SMSplex boundaries. This worked, but since the HSM migrations that "emptied" the dump volumes took place in another SMSplex, the devl sysplex's SMS (COMMDS) didn't have an accurate picture of free space. This is probably almost a worst case scenario for sharing between SMSplexes because "huge" files were created in one plex and deleted in another (as opposed to LOGR or "every day" allocations from a batch pool). But regardless it worked well until I turned on the striping. To answer your question, the high threshold was something like 85 and ISTR the low being something like 75. I set them to 95/10 and I think I also changed "AUTO MIGRATE" to I. "When AUTO MIGRATE is I, migration is done when the space used exceeds the half way mark between the HIGH and LOW threshold. " But another factor here was the migration of all the volumes in the pool to 3390-9. The more free space on a volume and in the pool, the better for this sharing since the other SMSplex may not have an accurate picture of the volume at allocation time. My problems were due to large dumps filling up a mod-3 and then they would get migrated from the prod plex, but the devl plex didn't know that space was available again (unless a new allocation "fit" and that volume was selected. So once the volumes were all changed from mod-3 to mod-9 along with the threshold changes, there was enough wiggle room for the allocations to take place without running the job. My statement above about "worst case" is because even though I have never set up shared "batch" volumes across the SMSplexes, I would guess it would work better with lots of smaller allocations on large volumes happening often from each plex. Every time a file is allocated or deleted from a particular plex, that plex would get a current picture of that space on the volume. We have shared LOGR across SMSplexes from day 1 and never had any problems at all and this is probably why. >>I just looked and all volumes in the LOGR pool are enabled on >>both SMSplexes also. There is a small development plex where >>there is also 2 SMSplexes ... but it is only a single volume in >>that pool shared between the 2 plexes. [...] > >I'm hoping we can "fence" allocations by system, and keep truly shared >volumes (ENABLED everywhere) to one or zero. Operlog is the only "problem" >child, and depending on where we use it and what RACF can do for us, would >be the only reason for a globally ENABLED volume. > It worked fine that way here before. I think they are all enabled now across the SMSplexes because there are 4 3390-9 in that pool now since our last DASD migration. It used to be 3 times as many volumes so they were easier to split up between the 2 SMSplexes. Regards, Mark -- Mark Zelden - Zelden Consulting Services - z/OS, OS/390 and MVS mailto:mzel...@flash.net Mark's MVS Utilities: http://home.flash.net/~mzelden/mvsutil.html Systems Programming expert at http://expertanswercenter.techtarget.com/ ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html