Subject: Re: Huge (> 1TB) disk
To: None <email@example.com, firstname.lastname@example.org>
From: Lord Isildur <email@example.com>
Date: 05/14/2002 10:27:26
Heres a reference to the track-aligned extents thing i mentioned:
Its in the usenix FAST 2002 proceedings, entitled 'track-aligned extents:
matching access patterns to disk drive characteristics'. it can be
downloaded from http://www.pdl.cmu.edu/ftp/stray/traxtent_abs.html
even off-disk, it can pay off to know the geometry.
On Tue, 14 May 2002, Lord Isildur wrote:
> even modern scsi disks have such a concept, and they make very good use
> of it in disk firmware when scheduling operations..
> and even outside of disk firware, being aware of tracks is both possible
> and useful.. check out http://www.pdl.cmu.edu/Freeblock and
> http://www.pdl.cmu.edu/Dixtrac.. the track-aligned extents stuff isn't online
> atm it seems..
> now, ide disks are _way_ worse at trying to present the disk as an
> abstract vector of blocks with no performance impact from particular
> access patterns...
> On 14 May 2002, Perry E. Metzger wrote:
> > On Mon, May 13, 2002 at 11:51:52PM -0400, Lord Isildur wrote:
> > > for example, one of the very cheap things with a pretty good win is to
> > > avoid allocating blocks which span track boundaries, as a quick way to
> > Haven't seen a disk where you had any idea what a "track" actually was
> > in years. I suppose "Lord Isildur" doesn't know about ZBR.
> > For reference, on modern drives, "geometry" is entirely a fiction.
> > --
> > Perry E. Metzger firstname.lastname@example.org
> > --
> > NetBSD: The right OS for your embedded design. http://www.wasabisystems.com/