Tom Lane wrote:
> easy or cheap to get a measurement that isn't skewed by kernel caching
> behavior. (You need a test file significantly larger than RAM, and
> even then you'd better repeat the measurement quite a few times to see
> how much noise there is in it.)
I found a really fast way in Linux to flush the kernel cache and that is
to unmount the drive and then remount. Beats having to read though a
file > RAM everytime.