[2842] in linux-scsi channel archive

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

block sizes

daemon@ATHENA.MIT.EDU (John Jannotti)
Fri Nov 21 12:07:46 1997

Date: 	Fri, 21 Nov 1997 12:06:43 -0500
From: John Jannotti <jj@lcs.mit.edu>
To: linux-scsi@vger.rutgers.edu


 I'm interested in getting some reasonably high performance disk io
out of the Linux kernel.  While perusing sd.c and hd.c is *seems* that
block_read will never make requests to disk (ide or scsi) for more
than 1k.  Is that true?  Or am I just missing how this works?

 (It looks like both drivers fill their respective blksize_sizes vector
with 1024, that's what I'm going on.)

 If I'm right, am I wrong in guessing that this is a big problem for
fast contiguous disk io?

 Again, if I'm right, can anyone guess how hard it would be to allow
larger requests?  I think I need at least 4k blocks, but others have
told be I should have my eye on 64k.

  jj

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