Your options are to have a system call to give hints to the filesystem
layer (e.g. I'm doing a big sequential read, don't cache the blocks). NT
allows this, as does (I think) BSD. This obviously requires minor source
changes to badblocks et al. OR, you could try to guess an appropriate
strategy from within the kernel based on past user program behaviour.
This has a higher "neat-o" factor, but is more difficult to do correctly
and efficiently.
N.B. We've got to stay ahead of Microsoft: NT just introduced
scatter/gather functionality in service pack 2 for 4.0. See
http://www.microsoft.com/kb/articles/q160/6/06.htm
for full details. Yes, I know Linux has had this for years.
-----Original Message-----
From: owner-linux-kernel-outgoing
[SMTP:owner-linux-kernel-outgoing@vger.rutgers.edu]
Sent: Sunday, May 04, 1997 4:18 PM
To: dledford; colin
Cc: linux-kernel
Subject: Re: Badblocks and no free pages...
Is there a way to limit write behind buffers so more memory is available
for other uses. Allocating more than eg. 2Mbyte per device won't speed
things
up I think. Programs like "badblocks" or "mke2fs" will use all available
space for buffering.
Chel