Re: [Gnu-arch-users] Some issues

[Andrew Suffield]

On Wed, Jun 09, 2004 at 08:10:50PM +0200, Florian Weimer wrote:
> * Andrew Suffield:
> 
> >> > This is essentially wrong on all the interesting filesystems. It's
> >> > equivalent on most of them, and a little slower to use one large file
> >> > on NTFS and HFS+.
> >> 
> >> Huh?  What about seek overhead?
> >
> > What seek overhead? There is no strong relationship between "being in
> > the same file" and "being in the same region of the disk",
> 
> Wrong, modern file systems try to maintain that relationship.

Wrong, that is not a strong relationship.

> > and no strong repulsion between "being in different files" and
> > "being in the same region of the disk". These things are mostly
> > related to creation pattern, not file membership.
> 
> While there are some file systems which group files according to the
> directory they are stored in them, UNIX file systems can hardly do
> this (and tla's hard links would break it anyway).

Right, ext[23] cluster based on creation order, which works out pretty
well for tla. I don't know where you got this "directory" nonsense
from.

> You can compensate for part of this problem if you sort directory
> entries by inode number before accessing them, but it's still much
> worse than just reading a file sequentially.

This is false, and it's also irrelevant, since your proposal was to
read a file randomly.

-- 
  .''`.  ** Debian GNU/Linux ** | Andrew Suffield
 : :' :  http://www.debian.org/ |
 `. `'                          |
   `-             -><-          |

signature.asc
Description: Digital signature