Subject: Re: msdosfs performance problem
To: None <tech-kern@netbsd.org>
From: Michael van Elst <mlelstv@serpens.de>
List: tech-kern
Date: 11/27/2005 11:59:49
tonnerre@thundrix.ch (Tonnerre LOMBARD) writes:

>If you want to touch a cluster on the disk in any way, you have to walk
>almost their entire disk structure until finally you reach the cluster
>you want.

I strongly doubt that this is the problem. First, the disk structure
is cached and wouldn't have to be read all over again. Second, it
is unlikley that the disk structure is as large as the file written.

So if these two are related then each write of a data block would
read exactly one block of meta data.


>FAT, the FAT contains the cluster number of the root directory, the root
>directory contains the cluster number of the first cluster of the file,
>and from then on you have a linked list of clusters following each
>other. I can't say I'm surprised that this works so poorly.

I can understand that this would be a problem for locating a file
or traversing a directory structure, but not for writing a single
file.

-- 
-- 
                                Michael van Elst
Internet: mlelstv@serpens.de
                                "A potential Snark may lurk in every tree."