From owner-freebsd-scsi  Wed Aug 20 00:32:41 1997
Return-Path: <owner-freebsd-scsi>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id AAA05789
          for freebsd-scsi-outgoing; Wed, 20 Aug 1997 00:32:41 -0700 (PDT)
Received: from sendero-ppp.i-connect.net (sendero-ppp.i-Connect.Net [206.190.143.100])
          by hub.freebsd.org (8.8.5/8.8.5) with SMTP id AAA05766
          for <scsi@FreeBSD.ORG>; Wed, 20 Aug 1997 00:32:34 -0700 (PDT)
Received: (qmail 5013 invoked by uid 1000); 20 Aug 1997 07:32:51 -0000
Message-ID: <XFMail.970820003251.Shimon@i-Connect.Net>
X-Mailer: XFMail 1.2-alpha [p0] on FreeBSD
Content-Type: text/plain; charset=iso-8859-8
Content-Transfer-Encoding: 8bit
MIME-Version: 1.0
In-Reply-To: <Pine.BSF.3.95q.970817195747.28679B-100000@misery.sdf.com>
Date: Wed, 20 Aug 1997 00:32:51 -0700 (PDT)
Organization: Atlas Telecom
From: Simon Shapiro <Shimon@i-Connect.Net>
To: Tom Samplonius <tom@sdf.com>
Subject: Re: expected performance with DPT RAID-5...
Cc: scsi@FreeBSD.ORG, Curt Welch <curt@kcwc.com>
Sender: owner-freebsd-scsi@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk


Hi Tom Samplonius;  On 18-Aug-97 you wrote: 

...

>   "cd" to one of the RAID-5 file systems and do a "dd if=/dev/zero
> of=testfile bs=64k count=1000".  That will create a file called
> "testfile"
> 64MB in length (64k x 1000), and tell you how long it took to write it.
> 
>   Then reboot (the only way I know of clearing the FreeBSD file cache,
> and
> the DPT cache), and do a "dd if=testfile bs=64k of=/dev/null" to read the
> file back.

If you unmount the filesystems and make sure no other process has any of
the device open, the caches will be flushed.  The O/S must do so in case
of removable media.  The DPT firmware flushes caches in response to 
ALLOW MEDIA REMOVAL which is issued by sd.c in response to a close call.

>   On the write test I get about 3500000 bps, and the write test about
> 9200000 bps.  These results don't seem very good too me.

For a RAID-5?  Your WRITE is a bit (40%?) slow.  READ is about right.
Remember, these are sequential tests.  You may want to adjust the
read-ahead cache if this is your typical load (some people enjoy
transferring many zeros from and to the disk :-).

If you want a random test, pick up form my system st.c.  It will do random
seeks and either read, write, or read-modify-write and tell you how badly
the system behaves.

An interesting variation is to run, say 256 dd's and 1024 st's.
Then measure throughput for a given stream.

While (or soon after) the test is run do:

    (cd /dev;./MAKEDEV dpt0)

    echo -n "dump softc" > /dev/dpt0
    get_dpt /dev/dpt0

Then look at the output.  Yes, I will write a manpage and a front-end
to this pair.  Right now read the source in sys/dev/dpt/dpt_control.c
Will answer many of your questions.

Simon