[plug] dd speed estimates and /dev/hdx grepping

[Harry McNally]

On Wed, 22 Oct 2003 20:37:10 +0800 Craig Ringer
<craig at postnewspapers.com.au> wrote:

> I don't know exactly why, but larger block sizes /will/ make your disk 
> image operation more efficient, at least unless you have really funky 
> hardware.

James Brown is a hardware hacker ? Ow ow ow.

If the kernel uses magic called scatter gather lists then multiple (small)
requests might be lumped into large writes to the drives. I've worked on
VME SCSI controllers that use a ring buffer of disk I/O requests this way.
But given that your disk I/O on any useful filesystem probably involves
non-contiguous access, the dd I/O is a bit "artificial" compared to normal
use so mayhaps disk I/O isn't optimised for this. Given that I don't know
the internals of the 2.4 kernel disk I/O this is all late night
speculation. Praps my first paragraph makes more sense and I should
sleep ;-)

The larger blocks that Craig has described would help when disks have
large caches and can suit themselves what and when to burst onto the
media as the heads whiz over it. The IDE controller writes a large block
which hits the drive cache and the rest is up to the drive. That would
help explain better performance.

SCSI also adds DISCONNECT and RECONNECT commands which even out-funks
Mr Brown http://home.capecod.net/~ccunning/funk.htm  

Harry
_______________________________________________
plug mailing list
plug at plug.linux.org.au
http://mail.plug.linux.org.au/cgi-bin/mailman/listinfo/plug