> For a page data set on a HiperPAV device, ASM creates two > sets of channel program control blocks. If a single page read > request come along while the first set is in use, ASM will use > the second set to start another channel program to do the single > page read. The purpose is to avoid having a single page read need > to wait for a long string of pageouts to complete before it > can be started.
>Jim, that's what you told me some time ago - and what I have presented in >Memory Matters: 2 effective per page data set. When I present on the >support for > 4GB page data sets I verbalise the trade off between the >simplicity of fewer, larger, page data sets vs the larger number of >channel program control blocks for more, smaller, page data sets. Because >of the virtualisation within modern disk controllers robustness favours >more, smaller. So to summarize: Page data sets can be up to 44,9GB in size (supported by ASM in using it), HiperPAV is used, but 'only' two IOs will ever get started for the same page data set at the same time. Jim, thanks for clarifying the current state. Martin, that means I have to attend the latest 'Memory matters' in May, right? :-) The reason I asked is that the DB2 colleagues have demanded an (almost) free-for-all memlimit for all address spaces. (Due to a clear abend0c4-4 on two getmained addresses in a BMC utility that they say is caused by not-enough-storage above the bar.) They will get 32GB per asid, which in turn means that we will now see drastic increase in paging, which was bad here to begin with (around 30% per page data set already). So we system guys need to redesign our paging setup, hence my question after studying ibmmain. For the curious: Doubling the memlimit from 500MB to 1GB increased paging to the same number of page data sets from 18-22% to 28-31%. Now we won't catch those applications anymore that just get 20GB, touch every page (which gets them backed) and then never use it because there just isn't any activity in the product on that system. But the locals will surely get filled up. Barbara ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@bama.ua.edu with the message: INFO IBM-MAIN
