If it is true that the dominant cost of paging is in seeks (and
that seems likely) then a compressed page cache is a good idea;
it's easy to compress and uncompress a page many times faster
than a seek.
BTW, does anybody have suggestions for how to do precise timings
inside or outside of the kernel? One of my students wrote some
routines that read the Pentium-family instruction-cycle counter,
so that we can time things down to a few nanoseconds. (This
is how we measure the speed of our compression algorithms in
userland.) If there are patches to do this in a better way, though,
I'd be very interested. (Or if somebody wanted to use my student's
code to implement hi-res time as a patch---maybe implementing
Solaris-like gethrtime---Mark gives it away free.)
Are there any plans to support hi-res time in the kernel? I think
all the new processors have cycle timers, and that sort of thing
is very handy for measuring the performance of system code.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/