Date: Wed, 13 Oct 1999 11:31:20 -0400 (EDT) From: Doug Ledford <dledford@redhat.com> To: Neil Conway <nconway.list@ukaea.org.uk> Cc: "P.C. Uiterlinden" <puiterl@hacom.nl>, "aic7xxx@FreeBSD.ORG" <aic7xxx@FreeBSD.ORG> Subject: Re: Tagged queueing on Dell WS610 - good idea? Message-ID: <Pine.LNX.3.96.991013112642.15910A-100000@devserv.devel.redhat.com> In-Reply-To: <99Oct13.091155bst.66305@gateway.ukaea.org.uk>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 13 Oct 1999, Neil Conway wrote:
> Thanks for the reply - I take it then that it is at least considered
> safe? That's reassuring. How about changing tag-depth without a reboot
> - possible? Or should I just hack the driver to do that ;-) Perhaps
> Doug can comment...
TCQ is considered safe. I've seen multiple reports of the system being
slower with deeper queue depths and I haven't had the time to look into
that at all. However, I suspect part of the cause for that is that the
low level driver actually tells the mid level code that the queue depth
for each device is 3 when TCQ is disabled. Internally in my driver then,
I let all 3 commands come down into the waiting queue and send the first
one out. When the first one completes, I send the second one to the card
immediately. We don't wait for lots of bottom end processing to take
place before the next command goes out. I'm certain that something that
would hurt untagged performance drastically is if I turned off that "lie
and tell them we can do 3 commands" option and made the first command
complete all the way back up to the upper layers before the next one got
sent. Many thanks to Leonard Zubkoff for providing the tip on doing it
this way with his BusLogic driver.
> thanks
> Neil
>
>
> P.C. Uiterlinden wrote:
> >
> > Neil,
> >
> > The best way to find out is just to try it. Use bonnie to measure the throughput
> > with different queue depths. In stead of bonnie, you also could use dd to write
> > and read a large file. Make sure the file is large in either case (much larger
> > than your RAM).
> >
> > Bonnie and dd gave comparable results in my case.
> >
> > To my astonishment, it turned out that setting the queue depth larger than 2,
> > severely impacted throughput with my disks.
> > [SNIP]
>
>
> To Unsubscribe: send mail to majordomo@FreeBSD.org
> with "unsubscribe aic7xxx" in the body of the message
>
Doug Ledford <dledford@redhat.com>
Opinions expressed are my own, but
they should be everybody's.
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe aic7xxx" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.LNX.3.96.991013112642.15910A-100000>