[9185] in linux-scsi channel archive
Re: Benchmarking and correctness
daemon@ATHENA.MIT.EDU (Brian Pomerantz)
Thu Jul 13 13:16:18 2000
Date: Thu, 13 Jul 2000 10:02:27 -0700
From: Brian Pomerantz <bapper@piratehaven.org>
To: linux-scsi@vger.rutgers.edu
Message-ID: <20000713100227.A25305@skull.piratehaven.org>
Mail-Followup-To: linux-scsi@vger.rutgers.edu
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
In-Reply-To: <Pine.LNX.4.21.0007131135040.27120-100000@challenge.dk>
On Thu, Jul 13, 2000 at 11:40:20AM +0200, Mads Bondo Dydensborg wrote:
> Hi there - a couple of questions I hope you will help me with
>
> I have made some modifications to a scsi driver, and need to benchmark
> it. I know about bonnie, but I was wondering if there where other relevant
> benchmarking utilities?
>
> Is there a way to flush the buffer cache? Right now I reboot to make sure
> I read from the disk, but that is kinda annoying.
>
> Also, are there any "correctness" tests? I have a disk connected to the
> SCSI controller and would like to be sure that I handle all the
> "neccesary" SCSI commands.
>
Using a raw device is a good way to get true numbers for your device.
Even a 'time dd if=/dev/raw1 of=/dev/null' style test can give pretty
good results especially if you want to know what different block sizes
will do for you. I was able to get really good results with this type
of testing to fine-tune the QLogic driver for my RAID hardware. If
you use a raw device, whatever benchmark you use will have to make
sure its buffers are aligned on a 512 byte boundary. Replacing calls
to malloc with calls to valloc is a good way to do this. It page
aligns your buffer which is even better :).
BAPper
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu
