From owner-freebsd-current  Tue Sep  9 05:25:52 1997
Return-Path: <owner-freebsd-current>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.7/8.8.7) id FAA04967
          for current-outgoing; Tue, 9 Sep 1997 05:25:52 -0700 (PDT)
Received: from smtp.algonet.se (tomei.algonet.se [194.213.74.114])
          by hub.freebsd.org (8.8.7/8.8.7) with SMTP id FAA04959
          for <current@FreeBSD.ORG>; Tue, 9 Sep 1997 05:25:49 -0700 (PDT)
Received: (qmail 27396 invoked from network); 9 Sep 1997 12:25:43 -0000
Received: from kairos.algonet.se (HELO kairos) (mal@194.213.74.18)
  by tomei.algonet.se with SMTP; 9 Sep 1997 12:25:43 -0000
Received: (mal@localhost) by kairos (SMI-8.6/8.6.12) id OAA14331; Tue, 9 Sep 1997 14:25:42 +0200
Date: Tue, 9 Sep 1997 14:25:42 +0200
Message-Id: <199709091225.OAA14331@kairos>
From: Mats Lofkvist <mal@algonet.se>
To: tlambert@primenet.com
CC: current@FreeBSD.ORG
In-reply-to: <199709090206.TAA00939@usr07.primenet.com> (message from Terry
	Lambert on Tue, 9 Sep 1997 02:06:08 +0000 (GMT))
Subject: Re: lousy disk perf. under cpu load (was IDE vs SCSI)
Sender: owner-freebsd-current@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

   It is unclear to me whether you are having problems on the disk on the
   NCR controller or the Adaptec controller.

Both, I got limited to ~20 user process xfers/s under load on all devices
including the single disk on the motherboard adaptec.

   The fact that you are running 2.2.2 does not fill me with hope; I was
   under the impression that 2.2.2 did not support tagged command queueing
   on the NCR, at least, and perhaps not the Adaptec as well.

   This would mean that you could have only a single concurrent operation
   outstandaing at one time -- which would certaily explain the performance
   problems you are seeing.

But can lack of tagged command queueing explain the combination of good
performance unloaded and bad with one or more cpu bound processes?

   You will have to check the list archives, but I believe there were also
   a number of issues with the 53c810a.  I have several 53c810b's, so I
   can't test for you.

I have no problem with issues with the 810a affecting devices connected
to it, but is it probable it could affect devices connected to the adaptec?

      _
Mats Lofkvist
mal@algonet.se


PS How do I find out if tagged queueing is supported? Maybe time
   to try -current?