OSEC

Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com
 
Subject: Re: SCSI, U2W in particular?, and poor performance...
From: Wietse Venema (wietseporcupine.org)
Date: Sat Mar 04 2000 - 21:53:36 CST

Well let it be said that it is not possible to safely (*) create
1024 files in 2 seconds on today's IDE disks with a non-journaling
file system. Doing so requires a multiple of 1024 disk updates in
different places (directory blocks, file blocks, inode blocks),
and disks just ain't that fast. At 10kRPM you have 3ms average
rotational latency. You can do the math.

        Wietse

(*) safely as in: should someone pull the plug the files must not
be lost. In other words, Linux may have told you that the files
are on disk, but in reality they aren't. Use a real file system
for Postfix if you care about reliability.

Justin Robertson:
>
>
> Linux Redhat, Linux Debian, Linux Slackware.
> All had the same issues... kernel version 2.2.12, 2.2.14, 2.2.14...
> The box I was talking about comparing it to was running Linux Debian w/
> kernel 2.2.14.
>
>
> Justin Robertson
> <zululinux.com>
>
> On Sat, 4 Mar 2000, Wietse Venema wrote:
>
> > Um, what SOFTWARE was runing on these boxen? There are
> > significant differences in speed and reliability.
> >
> > Wietse
> >
> > Justin Robertson:
> > >
> > > Using fsstone and smtpstone, I've noticed some 'quirks' in performance.
> > > For example, on an extremely low end Celeron 400 with 32MB of ram and a
> > > cheap 6.4gig IDE HD, I can succesfully create/delete 1024 files using
> > > fsstone in about 2 seconds. Using smtp-source I can send 1000 15k messages
> > > in 44 seconds, using only 2% of the CPU. Now, this is good, however, when
> > > the same tests are attempted on a conciderably higher end box, 800MHz,
> > > 768MB of RAM, and 10,000rpm 18.2gig U2W SCSI HDs, the results are... well,
> > > far lower then the low end machine. It takes 14 seconds to perform the
> > > same fsstone test, and well over 7 minutes to perform the same
> > > smtp-source test. Perplexed I attempted a series of CPU and HD tests on
> > > the high end machine, and it was performing as would be expected. Is there
> > > something about the way that postfix reads/writes data that is 'iffy' in
> > > the SCSI drivers at this point? This would make no sense to me... I've
> > > check with another individual whom is using a different SCSI controller
> > > card, ncr xxx vs. my adaptec a2940u2w, and the results are comprable. What
> > > gives? Anybody have an idea? I could just put everything on to a RAM disk,
> > > but the point is that a crappy machine shoudln't be smoking the highend
> > > box.
> > >
> > >
> > > Justin Robertson
> > > <zululinux.com>
> > >
> > >
> > >
> > >
> > >
> > >
> >
> >
>
>
>