Re: Support for 4KB sectors size disk ?

[David Holland <dholland-current%netbsd.org@localhost>]

On Tue, Jan 12, 2010 at 06:37:20PM +0000, David Laight wrote:
 > On Tue, Jan 12, 2010 at 05:04:44AM +0000, David Holland wrote:
 > > On Mon, Jan 11, 2010 at 10:07:42PM -0500, Thor Lancelot Simon wrote:
 > > > No, they don't present 512 byte sectors to the host OS.  They present
 > > > 512-byte sectors to *applcation code* running on the host -- this is
 > > > because the host OS uses the additional space itself.  The
 > > > sectors presented over the SCSI interface to the host are 524 bytes
 > > > or, in some older designs, 528 bytes.  Yank a disk out of a storage
 > > > array on one of these machines and see for yourself!
 > >
 > > ...and this is something we ought to be able to support going forward,
 > > if we can swing it.
 > 
 > I'm not sure that support for 'non power of 2' drives (below the disk
 > driver itself) is particularly useful.

Isn't it? You need fs-level support, but there are plenty of ways in
which a few bytes of per-block metadata can buy you a lot. Suppose for
example that FFS were extended so each block carried with it the inode
number and offset it belonged to; this would make volume resize vastly
easier and would eliminate the wrong-contents-after-crash bug without
needing softupdates. You could also store checksums, readahead hints,
transaction IDs, etc., etc.

It's not an accident that these drives are found in high-end
storage servers.

-- 
David A. Holland
dholland%netbsd.org@localhost