Date: Fri, 12 Mar 1999 15:12:25 -0800 From: Darryl Okahata <darrylo@sr.hp.com> To: freebsd-scsi@FreeBSD.ORG Subject: Re: SCSI disk oddities and Buslogic problems.. Message-ID: <199903122312.PAA12555@mina.sr.hp.com> In-Reply-To: Your message of "Fri, 12 Mar 1999 09:43:00 %2B0100." <3236.921228180@verdi.nethelp.no>
next in thread | previous in thread | raw e-mail | index | archive | help
sthaug@nethelp.no wrote:
> Chris Csanady <cc@137.org> wrote:
> > In this case, I have found that while running the following commands that
> > the data transfer is very erratic.
> >
> > dd if=/dev/rda2s3 of=/dev/null bs=128k (or any place on the disk..)
> >
> > iostat -w 1 da2
>
> No problem with a 9 GB model here. Typical iostat output:
I have seen strange things with 3.1-RELEASE (I've yet to try
-current), and tagged queueing. Using 3.1-RELEASE, an Adaptec 7895
controller (built into my Gigabyte 6BXDS motherboard) and two IBM
Ultrastar 9ES drives striped together using vinum, I get strange
performance numbers using iozone (yeah, iozone isn't very good for
benchmarking striped drives, but I'm just using it as a zero-th order
gauge).
Individually (if I do not use vinum and just newfs a single drive),
I get around 11MB/s write, and 13MB/s read. No problem here.
If I stripe the drives together using vinum, and use the default
newfs values, I get something like 3.7MB/s write, and 16MB/s read. The
write throughput sucks. If I run newfs with 4K frag sizes, the write
throughput jumps up to around 7MB/sec. Better, but still not great.
However, if I go into the BIOS and disable SCSI disconnect, the
write throughput jumps back up to 11MB/s. Much better, but disabling
disconnects kinda defeats the purpose of striping ....
The only thing I can think of is that tags are somehow involved, as
disabling disconnects also disables tags.
However, one slightly unusual thing about the striped drives is
that one drive is a plain fast/wide disk ("IBM DDRS-39130W S97B"), and
the other is the LVD version running in single-ended mode ("IBM
DDRS-39130D DC1B"), but this shouldn't matter.
--
Darryl Okahata
darrylo@sr.hp.com
DISCLAIMER: this message is the author's personal opinion and does not
constitute the support, opinion, or policy of Hewlett-Packard, or of the
little green men that have been following him all day.
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-scsi" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199903122312.PAA12555>