Date: Fri, 1 Aug 1997 23:52:39 +0200 From: Stefan Esser <se@FreeBSD.ORG> To: Michael Smith <msmith@atrad.adelaide.edu.au> Cc: "Christopher R. Bowman" <crb@glue.umd.edu>, dec@phoenix.its.rpi.edu, questions@FreeBSD.ORG, hardware@FreeBSD.ORG, Stefan Esser <se@FreeBSD.ORG> Subject: Re: SCSI Card to use. Message-ID: <19970801235239.39970@mi.uni-koeln.de> In-Reply-To: <199708010532.PAA03730@genesis.atrad.adelaide.edu.au>; from Michael Smith on Fri, Aug 01, 1997 at 03:02:16PM %2B0930 References: <Pine.SOL.3.95q.970801012004.5842A-100000@drpepper.eng.umd.edu> <199708010532.PAA03730@genesis.atrad.adelaide.edu.au>
next in thread | previous in thread | raw e-mail | index | archive | help
On Aug 1, Michael Smith <msmith@atrad.adelaide.edu.au> wrote:
> Christopher R. Bowman stands accused of saying:
> >
> > would have made this suggestion myself, however I don't believe that
> > the NCR driver is doing Ultra-Wide transactions yet is it? The probe of
> > my IBM Ultrastar 2ES Ultra-Wide drive (included below) indicates that
> > I am not getting these types of transfers with my 2.2.2 (from the CD)
> > system. Thus I didn't suggest 875 based cards since the request asked
> > for Ultrawide cards and though the my Tekram 390F is indeed an ultrawide
> > 875 based card which I am QUITE hapy with, it is not doing Ultrawide
> > transactions.
> >
> > Perhaps this is not the case with newer source(I don't know) but people
> > should be aware of this before purchase.
>
> Code for this was committed recently. Pester Stefan (se@freebsd.org) if it
> hasn't made it into 2.2-stable yet.
Support of 53c875 based cards was important to me, and I spent
some effort in their support at the end of last year. I did not
have time to add support for extended features (like Ultra-SCSI
transfers, or the on-chip SRAM of the 825A and 875), and in fact
wanted to be sure, that "normal" speed transfers were working,
first.
(I had quite a hard time some years back, when many hardware
problems were assumed to be driver bugs, most were not, but
instead caused by hardware problems, often bad cables and
terminators. Ultra-SCSI demands much higher cable quality again,
or much shorter SCSI bus length, and since I had been unable to
distinguish driver problems from hardware problems, I decided
to not go for Ultra.)
But the lack of Ultra support in the NCR driver has really lasted
for too long, and for that reason I was glad when Gerard Roudier
(again) offered to provide me with patches that he had worked out
over the last few months.
The code in -current contains his patches, and should fit into
2.2.x (and -stable) without problems (though I did not yet try
this myself). I will wait for a few more weeks, and will then
update the driver in -stable, too.
The patches show significant improvement of the (already good)
performance of the driver, as much as 10% in certain tests.
SCSI command overhead has again been reduced, and I received the
following Bonnie results from Gerard (who has a significantly
faster machine than my old 486 based SP3G :)
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
Atlas I 100 2436 50.4 6387 23.4 2846 18.4 2729 55.5 6576 19.3 98.0 3.8
Atlas II 100 2385 50.0 8721 39.2 3578 25.9 2767 56.2 9205 38.6 98.7 4.6
(Numbers are for a -current on a P133, 512KB cache, 32MB EDO with a
WIDE Atlas and Ultra-WIDE Atlas II, I'm not sure about the zones the
file systems resided in ...).
If you think these numbers are low:
They were measured by running two Bonnie processes simultanously,
which also explains the 50% of the CPU each process got ...
A total of 15MB/s was written and 16MB/s read from both drives
combined. (The tests diverged since the Atlas II was significantly
faster than the Atlas (I), and this explains a CPU load of some 55%
each for the "read char" test ...)
Regards, STefan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19970801235239.39970>