Memory Pages/sec and Pages Input/Sec thresholds

Aug 24, 2012 at 12:13 AM

Hi, I am confused by the threshold of 1000 that is used by PAL for both the Memory Pages/sec and Pages Input/Sec counters. All the research that I have done on this indicates that the threshold should be closer to 20 or 25. Why the big discrepanancy or am I missing something?




Aug 26, 2012 at 5:06 AM

Hi Cjells,

\Memory\Pages/sec and related counters like Page Inputs/sec measure hard page faults to and from the disk. x86 and x64 Windows computers use a default 4 KB page size by default. If you have a threshold of 20, then you are saying that the disk is not able to handle 80 KB per second (4 KB x 20). Most 7200 RPM disk drives can handle about 10 MB per second. 4 KB x 1000 is 4 MB, so a 7200 RPM disk drive should be able to easily handle 4 MB per second which is the equivalent of 1000 hard page faults per second. This is why I start PAL off with 1000 and why everyone else should update their thresholds to be in line with modern hardware.

In any case, you should never use pages/sec by itself because the hard page faults could simply be normal disk access. Many people think that hard page faults only refer to the paging file, but that isn't true. I can easily get a burst of 1000 pages/sec just by opening up Word on a system with no paging file. Backup software typically uses memory mapped files which generate hard page faults (none of which are related to the paging file). With that said, page file access is included in pages/sec, but you have to consider all of the other types of hard page faults.