page reads/sec vs. disk reads/sec

[Bob]

From the Explain text in performance monitor, memory\page
reads/sec is a measure of the number of times per second
that pages were read from disk. Physical disk\disk
reads/sec is the rate of disk reads (presumably all disk
reads). Why, then, are there fewer disk reads/sec on the
single physical disk that holds the paging file than there
are page reads/sec when both are displayed with a scale of
1.000? Disk reads should always be equal to or greater
than page reads, yes?

Thx,
Bob

 

[Dan Seur]

Bob - this is speculation, but I suspect it's hardware that discovers
that a requested disk access is already in store in the pagefile. If
this is true, and such pagefile "hits" are not counted as real disk
accesses, then the counters are counting different events, neither is a
sub- or super-set of the other, and the apparent anomaly is explained.

Perhaps someone with better knowledge of the latest OS/chip structures
can confirm or set me straight. Keeping track via hardware stacks of
what disk clusters were already in store was a basic S/360 and early x86
architectural 'performance' feature and has I think been much enhanced
since then.

If there are fewer real than virtual disk accesses on your system,
you've got a system that's humming right along! Congrats.