httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From dean gaudet <>
Subject Re: mod_file_cache performance
Date Mon, 02 Jul 2001 23:23:49 GMT
fyi -- the existing zero-copy capabilities of linux 2.4.x (x>=5 i think,
but probably also included in the redhay 7.1 kernel along with the TUX
stuff) are triggerred only if you use sendfile() ... nobody has done the
work to enable them for mmap().  so whatever's needed to activate
sendfile()  should be a win provided you've got the right hardware.

for 100baseT, "the right hardware" is basically the 3com 3c905b or 3c905c.
for gig-e you can use the acenic (which is now available in the netgear
GA620 and 620T, and some 3com card i forget the card#) or syskonnect (i
think).  a handful of other drivers are still under development.

anyhow, to see the difference with zero-copy enabled you'll probably need
to study idle cpu (especially if it's only 100baseT).


On Mon, 2 Jul 2001, Cliff Woolley wrote:

> On Mon, 2 Jul 2001, dean gaudet wrote:
> > is this across loopback?
> >
> > if so then it's useless for performance tuning/measurement.  you gotta use
> > a real network, and a beefy enough client to get any meaningful results.
> I assumed that would make a difference... I just wanted to make sure we
> weren't doing something *ridiculously* stupid.  =-)
> > that said, you might want to use lmbench <>
> > to study what file open times are like under linux.  they're pretty damn
> > small.  linux's dcache kicks a lot of ass... and in the end, userland
> > caching of stuff like this just means you're duplicating efforts.  that
> > means a waste of L2, which will have a non-linear decrease on your
> > performance.
> I'd wondered if that might not be the case.  It makes sense.  I'm in the
> middle of running my tests again, and give or take a margin of error, it's
> pretty much exactly the same regardless of whether you use the cachefile
> directive or not.  MMapFile speeds things up a tad, but not much.  I'll
> post those (also useless ;-) numbers a little later.
> --Cliff
> --------------------------------------------------------------
>    Cliff Woolley
>    Charlottesville, VA

View raw message