systems without pread?
jpeach at sgi.com
Fri Dec 19 05:22:37 GMT 2003
On Fri, 19 Dec 2003 09:40:59 +1100
James Peach <jpeach at sgi.com> wrote:
> On Tue, 16 Dec 2003 15:12:21 -0800
> Jeremy Allison <jra at samba.org> wrote:
> > On Wed, Dec 17, 2003 at 09:30:22AM +1100, James Peach wrote:
> > > >
> > > > Hmmmm. How have you tested this performance win ?
> > >
> > > I have a smallish test setup using an Origin 3k server and 4 Origin
> > > 300 clients. I'm connecting each client to the server through a
> > > different gig-e NIC and running multiple load generators on each
> > > client. I'm gathering performance numbers using Performance Co-Pilot -
> > > these include NIC throughput, CPU time, samba IOPs per second, and a
> > > bunch more stuff.
> > What does the performance change look like on Linux ? Have you been
> > able to do tests there yet ?
> A preliminary experiment on write throughput (single client) running on
> Linux 2.4.21 on IA64:
> Workload: Streaming Write (pwrite)
> threads network.interface.bytes
> 16384 12 82.45 (in) 1.58 (out) (with pread)
> 16384 12 75.99 (in) 1.41 (out) (without pread)
Ok, I have a reasonable amount of data now. It looks like using pread/pwrite
is not as big a win on Linux as it is on IRIX. There is a gain of 5% - 10%
in write throughput for low number of clients (<24), but the read
performance is not significantly different when using pread (+/- a couple of
%). Metadata workloads (ie. netbench) show no difference. My feeling is that
if you are running Samba on a platform with (relatively) expensive system
calls, this will help, otherwise it's pretty neutral.
James Peach | jpeach at sgi.com | SGI Australian Software Group
I don't speak for SGI.
More information about the samba-technical