Cached file read performance with 6.2-PRERELEASE

Mark Kirkwood markir at
Thu Dec 21 15:29:14 PST 2006

Oliver Fromme wrote:
> Mark Kirkwood wrote:
>  > Exactly, that's why I did the comparison - I think you missed the part 
>  > where I mentioned the 2 systems were *identical* with respect to cpus, 
>  > memory, mobo - in fact even the power supplies are identical too!
> So I assume your benchmark measured the performance of the
> zero and null devices under FreeBSD and Linux.

No - that was peripheral to the benchmark, and I should not have sent
that message 'cause actually I've taken dev/zero and /dev/null *out* of
the picture - check earlier messages with the .c prog attached, I'm
using read(2) and lseek(2) to access a "real" file, that just happens
(i.e. has been arranged) to be cached!

> This is a quote from the "cstream" docs:  "These special
> devices speed varies greatly  among operating systems,
> redirecting from it isn't appropriate  benchmarking and
> a waste of resources anyway."
> I suggest you try cstream (ports/misc/cstream) instead of
> dd.  It supports built-in zero creation and data sink, so
> you don't have to use the zero and null devices at all,
> eliminating their overhead.  It would be interesting how
> that will change your benchmark numbers.

Thanks - I was suspicious of these special files, but had no evidence!



More information about the freebsd-stable mailing list