On Tue, 2005-11-01 at 18:49 +0000, Peter Grandi wrote:
> Hi, I have recently (six weeks ago) switched all the filesystems
> on my PC from 'ext3' to JFS for various reasons, in part because
> I like the JFS design, in part because I had discovered that
> over the past several months my 'ext3' ''root'' filesystem
> performance had degraded by 7 times over that of a freshly
> loaded copy of itself.
> 
> So I have decided to compare the read speed of my ''root'' JFS
> filesystem as it is now, after six weeks of in-place package
> upgrades, and how it would be if I reloaded it.

Great.  I have never had this kind of data.

> I have an otherwise quiescent disc, and I copied my ''root''
> filesystem to it first as a partition image, to preserve the
> ''used'' layout, and then by 'tar', so that it would be reloaded
> in an optimal ''new'' layout. The filesystem contains around
> 7.5GiB of data in 360k files, and 2.3GiB are free.
> 
> The result is that a whole-filesystem 'tar c' on the ''new''
> layout takes 10min., on the ''used'' layout takes 26min., which
> is a factor of over 2.5 times longer. I have also done some spot
> checks on some largish files that I know have been regularly
> updated/rewritten, and there are similar slowdowns (no slowdown
> on files that have not been updated in the past six weeks).
> 
>   Details here: http://WWW.sabi.co.UK/Notes/anno05-4th.html#051101

This makes a great case for implementing a defragmenter.  I'd put it off
in the past for a few reasons.  1) there were always more urgent things
that needed to be done, 2) the defrag tool that ran on OS/2 was very
limited in what it did, and I didn't think that porting it alone would
be sufficient, 3) I didn't have any data to demonstrate that
fragmentation was a problem.

(I noticed the comment about sync adding 4 minutes due to the modified
atimes.  Have you considered mounting with the noatime option?)

> Now 2.5 times is a lot better than 7 times (but the latter was
> over a rather longer period on a filesystem with less free space),
> but it is still somewhat disappointing, as the average transfer
> rate for reading the whole filesystem goes down (on a disc
> capable of around 35MiB/s sustained in optimal single-large-file
> conditions) from 12MiB/s, which is reasonable, to 5MiB/s, which
> is not awesome.

Agreed.

I still don't anticipate having the time to work on a defragmenter in
the near future.  This would be a good project from someone wanting to
contribute something to jfs.  I'd be happy to help.

-- 
David Kleikamp
IBM Linux Technology Center



-------------------------------------------------------
SF.Net email is sponsored by:
Tame your development challenges with Apache's Geronimo App Server. Download
it for free - -and be entered to win a 42" plasma tv or your very own
Sony(tm)PSP.  Click here to play: http://sourceforge.net/geronimo.php
_______________________________________________
Jfs-discussion mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/jfs-discussion

Reply via email to