Could you please let me know if there is any threshold value above which 'average page faults/second' should be considered a serious issue?
Most materials say that 'high value' of page fault combined with low unreferenced interval count means the memory is under a lot of stress, but is there any range to look out for, like 10's or 100's or 1000 page faults/second.
Occasionally we get 4 to 7 average faults/second. is this something to be worried about.
(Don't worry that someone put a mainframe on a novice's hands who does not even know anything about performance :-) I have access to SMF and allowed to play with it, so poking around)
Joined: 06 Jun 2008 Posts: 8449 Location: Dubuque, Iowa, USA
Personally, I think it will vary. 10 page faults per second in a CICS region implies 10 transactions (or maybe 1 transaction 10 times?) per second are getting stopped while paging occurs. If you're running 2000 transactions per second, probably no big deal (except to the users affected) but if you're running 100 transactions per second then somebody ought to be looking closer.
z/OS is designed to tolerate high paging rates, but application performance will suffer the more paging occurs. Occasional periods of 4 to 7 per second is probably just a spike in memory usage, but 4 to 7 per second from 9 am to 5 PM (or 11 PM to 5 AM) might be a different story.
If you're running 2000 transactions per second, probably no big deal (except to the users affected) but if you're running 100 transactions per second
4 to 7 per second from 9 am to 5 PM (or 11 PM to 5 AM) might be a different story.
Thanks for this perspective. Gives some idea of how to look at this measure and get meaning out of it. We have approximately 70 IMS transactions/sec and 4-7 page faults calculates to approximately 10% of the transactions on some days. Guess its worth analyzing what is causing it.