Last time I looked at PDSE performance was before PDSE V2. And I checked specifically PDSEs with RECFM=VB. About 10000 member in a 18000cyl PDSE library. ISPF 3.4 took about 90 seconds. That time was spent on I/O (reponse time about 4ms for each I/O). During that time about 10000 I/Os were done to that data set. AFAIK, the 'directory' of a PDSE is strewn in with the actual data (PDSEs used to use Media Manager Code for I/O, i.e. 4K blocks for each I/O). Adding the response time for the I/O more or less resulted in the 90 seconds wait time.
I believe that inserting an entry is just finding the place where it belongs and adjusting the previous and next pointer(s). As for HFS - in a former life we ran Lotus Notes on z/OS. Over time, performance in access to the HFS degraded. IBM recommended migrating to zFS. Bad Move. We ran into several zFS software problems and ended up going back to HFS. Lo and behold, performance was MUCH better on HFS then, probably because the migration back and forth had reorganized the underlying data structures. As for caching: The SMSPDSE1 address space used to cache the 4K blocks. Which did not help at all, because back then the maximum cache available to SMSPDSE1 was 16GB (I believe). We had about 10 of those large VB PDSE's, and together they were much bigger than the available cache, and SMSPDSE1 would cache the full 4K (which includes data). The nature of access was 'go search them all for a listing that fits', so it regularly took a long time since cache content in SMSPDSE1 got replaced. In production we ended up converting back to PDS because that was MUCH faster. I have no idea how PDSE V2 compares to my experience. If I ever find the time I might test on a VB PDSE. Barbara ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
