From owner-freebsd-hackers  Sat Oct  7 15:01:00 1995
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.6.12/8.6.6) id PAA26319
          for hackers-outgoing; Sat, 7 Oct 1995 15:01:00 -0700
Received: from aslan.cdrom.com (aslan.cdrom.com [192.216.223.142])
          by freefall.freebsd.org (8.6.12/8.6.6) with ESMTP id PAA26312
          for <hackers@freebsd.org>; Sat, 7 Oct 1995 15:00:53 -0700
Received: from localhost.cdrom.com (localhost.cdrom.com [127.0.0.1]) by aslan.cdrom.com (8.6.12/8.6.9) with SMTP id PAA26282; Sat, 7 Oct 1995 15:00:02 -0700
Message-Id: <199510072200.PAA26282@aslan.cdrom.com>
X-Authentication-Warning: aslan.cdrom.com: Host localhost.cdrom.com didn't use HELO protocol
To: Bruce Evans <bde@zeta.org.au>
cc: dennis@etinc.com, hackers@freebsd.org
Subject: Re: VLB Disk Controllers 
In-reply-to: Your message of "Sun, 08 Oct 1995 07:15:43 +1000."
             <199510072115.HAA03150@godzilla.zeta.org.au> 
Date: Sat, 07 Oct 1995 15:00:01 -0700
From: "Justin T. Gibbs" <gibbs@freefall.FreeBSD.org>
Sender: owner-hackers@freebsd.org
Precedence: bulk

>>Justin Gibbs says,
>
>>>>Adaptec, Buslogic. Any possitive or negative comments on the drivers would
>>>>be welcome.
>>>>
>>>>Dennis
>>>
>>>The 2842 seems to work well, supports tagged queuing and other advanced
>>>SCSI II features that the buslogic (may be a driver limitation) does not.
>>>The adaptec has much lower command overhead than the buslogic
>
>That's not saying much. An IDE controller has a much lower command
>overhead than the buslogic:

It also has much higher CPU overhead, but we've been down that road
before. :)

Adaptec 2742 (Rev C aic7770) on 486DX2/66 VL/EISA - Quantum PD1225:
Output for disklatency /dev/rsd0:
Command overhead is 2002 usec (time_4096 = 2056, time_8192 = 2111)
transfer speed is 7.50362e+07 bytes/sec

Adaptec 2742 (Rev C aic7770) on 486DX2/66 VL/EISA - Quantum Empire 2100:
Output for disklatency /dev/rsd1:
Command overhead is 1299 usec (time_4096 = 2754, time_8192 = 4208)
transfer speed is 2.81672e+06 bytes/sec

Neither of these drives even approach the command processing speed of
a Grand Prix (QUANTUM XP34301 that you tested), and I would guess that
the latency would be even lower for that drive.  I know the Grand Prix
is fast because it found all the race conditions in our sequencer code! :)

We also don't see the effects of tagged (ie overlapped) commands on the
SCSI bus.  Most of the overhead you see is diminished by the fact that
occurs while the drive is seeking to satisfy an earlier request.

>A high command overhead causes slow file system operations for
>everything except large i/o's.

Yup, but your whole machine is affected by a PIO device.

>Bruce

--
Justin T. Gibbs
===========================================
  Software Developer - Walnut Creek CDROM
  FreeBSD: Turning PCs into workstations
===========================================