[7851] in linux-scsi channel archive

home help back first fref pref prev next nref lref last post

Re: Performance question

daemon@ATHENA.MIT.EDU (Douglas Gilbert)
Tue Jan 18 03:51:46 2000

Message-ID: <3883C2AC.9AC26A3E@interlog.com>
Date:   Mon, 17 Jan 2000 20:32:28 -0500
From: Douglas Gilbert <dgilbert@interlog.com>
MIME-Version: 1.0
To: Achim Leubner <achim@vortex.de>
Cc: linux-scsi@vger.rutgers.edu, linux-kernel@vger.rutgers.edu
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

Achim Leubner wrote:
> 
> Hello,
> 
> I'm the maintainer of the ICP Disk Array Controller driver (gdth) and
> I have a performance question.
> I read a big file on an ext2 filesystem under Linux 2.2.xx with "dd"
> with a read block size of 1 MB.
> In the gdth scsi driver and therefore on the controller I see
> 1. read commands with a size of  <= 64 KB only
> 2. max. 2 simultaneous read commands only
> If the commands are splitted into 64 KB I/O's, why I don't see 16
> simultaneous commands (64 KB * 16 = 1 MB) ?
> Could the reason be the read ahead cache implemented in
> mm/filemap.c ?
> Is there any possibility to change any filesystem/kernel parameters
> to get I/O's with greater size or to get more simultaneous I/O's ?
> 
> Any hints are greatly appreciated.

Achim,
Have a look at the scsi generic (sg) web site at:
http://www.torque.net/sg  [follow utility + fast copy links]

In the utilities is a "dd" variant called sgp_dd. It allows 
the number of blocks per transfer (i.e. per SCSI command) to 
be controlled as well as the number of commands queued to
the device. The latter is done by POSIX worker threads, the
number of which are specified on the command line.

sgp_dd is significantly faster than dd performing disk to disk 
copies.

Doug Gilbert

-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu

home help back first fref pref prev next nref lref last post