From owner-freebsd-hackers  Wed May  1 20:00:39 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id UAA00908
          for hackers-outgoing; Wed, 1 May 1996 20:00:39 -0700 (PDT)
Received: from dyson.iquest.net (dyson.iquest.net [198.70.144.127])
          by freefall.freebsd.org (8.7.3/8.7.3) with ESMTP id UAA00881
          Wed, 1 May 1996 20:00:30 -0700 (PDT)
Received: (from root@localhost) by dyson.iquest.net (8.7.5/8.6.9) id CAA00185; Thu, 2 May 1996 02:55:42 -0500 (EST)
From: "John S. Dyson" <toor@dyson.iquest.net>
Message-Id: <199605020755.CAA00185@dyson.iquest.net>
Subject: Re: lmbench IDE anomaly
To: terry@lambert.org (Terry Lambert)
Date: Thu, 2 May 1996 02:55:41 -0500 (EST)
Cc: luigi@labinfo.iet.unipi.it, terry@lambert.org,
        msmith@atrad.adelaide.edu.au, koshy@india.hp.com, hackers@FreeBSD.ORG,
        current@FreeBSD.ORG
In-Reply-To: <199605020128.SAA10906@phaeton.artisoft.com> from "Terry Lambert" at May 1, 96 06:28:40 pm
X-Mailer: ELM [version 2.4 PL24 ME8]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: owner-hackers@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

> > > Anyway, the results showing SCSI being better than IDE are certainly
> > > valid.
> > 
> > So we are back to the regular "SCSI is better than IDE" debate...
> 
> "Better" as in a "lower system overhead unless you've written a PIO4
> EIDE driver that you haven't shared with the rest of us and overcome
> the interrupt bugs in 3 out of the 4 most popular IDE controller
> chipsets and overcome the PIO4 probe crashing non-PIO4 systems and
> even then, the DMA overhead is slightly higher than SCSI and EIDE
> CDROM's use SCSI commands over the IDE interface anyway" kind of way.
> 
> Purely "subjective".  ;-).
> 
> 
On FreeBSD, as long as your bios (at least on my ASUS TPN motherboard)
sets the drive mode to a good one, you get the advantage of the PIO mode
that the bios senses.  On my new 2.5GByte WD Caviar, I get 8-9MBytes
read perf on FreeBSD.  I have been looking at implementing DMA using Garrett's
IDE dma code, but haven't gotten to it yet.

The WD type drives don't/can't take advantage of things like tagged queuing,
and the 2.5Gbyte drive that I have only has a 128K cache.  But, I'll put
my 200usec or so command overhead against any AHA2940/Atlas drive anytime.
In some ways, the EIDEs are pretty impressive.

Bottom line, the EIDE drives do appear to be slower in many if not most
respects than a very good SCSI drive, but my 2.5Gbyte Caviar is almost
always faster than my 2Gbyte SCSI Hawk.  (With perhaps a little more
overhead -- but maybe not.)  With the 2.5Gbyte Caviar's being
about $400US in wholesale, with no SCSI adapter needed, they are probably
pretty darned good price/perf for workstation use.

John