Page Fault Rate

[oradba]

Hi,
We have a HP-UX 11.0 system showing following characteristics :

Page Faults ( Cummulative ) : 70.4
Page Ins ( Cummulative ) : 40.6

Do the above figure indicate a problem ?

thanks
oradba

[John Bolene]

If these are per second, you need more memory.

If these are just a count, it depends on when the counting started.

[Sridhar Bhaskarla]

Hi,

I would rename the 'page fault' to 'page miss' as people often get alarmed with the term fault.

page fault happens when a process tries to access code that is not yet mapped into memory. Then the system will page-in and get the pages.

You would need to be worried if you see any abnormal values in "po" column in 'vmstat'. Also, you would see a non-zero figure in the "KB Used" column of your "swapinfo " command for the device swaps. And both the above are indications of memory issues only when you have 95% or more memory utliization.

-Sri

[Michael Steele_2]

I agree with Mr. Bashkarla:

Don't worry about it but look for po in vmstat. This indicates a process was swapped out before completing its time slice.

Page outs are nowadays a mostly non-existent and rarely if ever seen problem because of the increased sizes in physical memory. In fact, it???s probably more common to have too much physical memory which is apt to cause system fragmentation with poorly tune kernel parameters. Monitor your sar -v outputs for some indication of this:

sar -v 5 5

???and note that you have a problem when the numerator and denominators become equal, or, the numerator is only about 10-15% of the denominator.

[Andy Zybert]

Page faults and pageins are meaningless if you are trying to identify a memory bottleneck.

A better indicator would be pageouts, but on HP systems should only be considered if memory usage is > 97%(see glance), there may also be a case to reduce the size of the filesystem buffer cache (normally defaults to 50% unless tuned via kernel param dbc_max_pct)which should be at most 300-500Mb.

The clearest indicator would be swapouts/sec (deactivations/reactivations). Any significant indication of swapouts would indicate either memory needs tuning, if poss, or more memory is required to manage the workload on the system.

Hope this has answered your query.

AndyZ