Re: Silly question, defrag

[Tracy R Reed <treed@ultraviolet.org>]

[-- Attachment #1: Type: text/plain, Size: 2379 bytes --]

On Thu, Apr 04, 2002 at 11:16:51AM +0200, Anders Widman wrote:
> I do not agree. I run a fileserver with a 814GB filesystem using
> ReiserFS (I have run NTFS and ext2/3 also). Modern filesystems might
> be smarter in storing new files by not packing them tightly.
> 
> In my case that workes fine up to a certain percentage, after that ALL
> new files are beeing fragmented due to the fact that there is only
> small blocks of space between all files. I don't see any filesystem
> that don't need defragmentation. Not in my case.

Yes, after a certain percentage you will start getting fragmentation. Will
you really notice the performance hit? Who knows. It depends on how much
and which files get fragmented. One solution is to not fill disks up to
more than 90%. That is what most people who have a need to worry about
such things do. I'm just saying that it isn't worth it. If you are at 90%
you need to buy more disk anyway because soon you will be at 100%. 

If there were real value in regularly defragging then Veritas, Sun, IBM,
HP, and all of those guys would have made defraggers for their respective
filesystems and it would be considered best practice and standard
operating procedure to use them. But I have never heard of any such tools
nor procedures.

This whole discussion results from the fact that so many Linux people come
from PC backgrounds where they were taught to habitually defrag. People
who come from other systems never give it a thought.

Nonetheless, I look forward to having this functionality is reiserfs
because it certainly can't hurt. What interests me even more than
defragging is performance optimized layouts. If the filesystem can somehow
keep track of patterns of frequently accessed blocks and could recognize
that one set of blocks on the inner cylinders is always read immediately
after reading a set of blocks on the outer cylinder (or perhaps instead of
keeping track of blocks which are read it would be more efficient to keep
track of commonly performed long seeks and move data to remove those) and
could rearrange things so that all of the needed data passes under the
read head in most often used sequence we would see a MUCH bigger
improvement in performance.

-- 
Tracy Reed      http://www.ultraviolet.org
Would you buy a car with the hood welded shut?
Linux: the maintainable OS.

[-- Attachment #2: Type: application/pgp-signature, Size: 240 bytes --]