Date: Wed, 9 Jul 2003 17:59:23 +0800
From: Federico Sevilla III (jijo@free.net.ph)
To: Philippine Linux Users Group Mailing List (plug@lists.q-linux.com)
Subject: Re: [plug] journaling file systems...
User-Agent: Mutt/1.5.4i

On Tue, Jul 08, 2003 at 07:45:51PM -0700, Carlos wrote:
> hi, i stopped using ext2 and started using reiser fs on all my linux
> installs.... but my question is, which is the best filesystem.. i know
> ext2 sucks compared to ext3, reiser fs etc....

I actively use ext3, ReiserFS and XFS on my GNU/Linux boxes. I have found that while all three are more than good enough for most applications, I prefer each for the following applications:

- ext3 for general conservative installations where filesystem speed will not be the bottleneck. Performance has increased with recent improvements, but in my non-scientific observation it still lags behind at least XFS for working on larger files. Reports say that ext3 with data=journal is great for mail spools, and directory indexing supposedly improves access for directories with many small files significantly.

I have found ext3 to be great for situations where the machines do not have backup power, as I have found it to be the most reliable with its data journalling or ordered-writes.

- ReiserFS has good overall performance but stands out when dealing with directories with many small files. I use it for our Squid cache, but it'll probably do great for Maildirs as well. Deletes are hideously fast. Be careful with them 'rm -rf /' commands. ;)

- XFS also has good overall performance, but stands out when dealing with large files. It used to be the only one with decent ACL support, but I believe ext3 has this, as well, and Reiser4 is supposed to take things even further with the plugin support. Older versions had some problems with files getting populated with binary nulls when written to immediately before a crash because of the way XFS handled its cache and preallocation of space. This has been fixed, though, and things are very stable now. Delete speeds are also much faster than before, when they used to be noticeably slow because they were done syncronously. They have a neat set of users profiled voluntarily in http://oss.sgi.com/projects/xfs/xfs_users.html.

> so what's the difference between xfs, jfs, reiser fs and ext3?? what
> are the pros and cons?? i googled for it on the net... but i also
> wanna hear some opinions in plug... would really appreciate it if you
> guys reply. thanks

If you want some antique comments, Orly Andico and I had a nice exchange of emails[1] circa 2001. I don't know how valid my flamboyant claims still are, if they ever were, but so that we don't have to rehash the thread, I'm giving you a link. There are other bits on journalling filesystems in the archives. Just be creative with your keywords. ;)

[1] http://marc.free.net.ph/thread/20010724.152952.95b5e3a4.html

--> Jijo

--
Federico Sevilla III : http://jijo.free.net.ph : When we speak of free
Network Administrator : The Leather Collection, Inc. : software we refer to
GnuPG Key ID : 0x93B746BE : freedom, not price.

Date: Thu, 10 Jul 2003 14:03:24 +0800
From: Federico Sevilla III (jijo@free.net.ph)
To: Philippine Linux Users Group Mailing List (plug@lists.q-linux.com)
Subject: Re: [plug] journaling file systems...

On Thu, Jul 10, 2003 at 09:23:50AM +0800, Gideon N. Guillen wrote:
> On Wed, 2003-07-09 at 17:59, Federico Sevilla III wrote:
> > behind at least XFS for working on larger files. Reports say that
> > ext3 with data=journal is great for mail spools, and directory
> > indexing supposedly improves access for directories with many
> > small
> > files significantly.
>
> Isn't this supposed to be "data=writeback". This one us supposed to
> have "highest-throughput" according to the mount man pages. :) Or is
> the "data=journal" the better for directories with many small files?

data=journal provides highest throughput for MAIL SPOOLS because the small files are written, sent, then removed. When this is all done in the journal things supposedly go much faster. Otherwise, data=writeback, which is the default, provides best overall performance AFAIK.

--> Jijo

--
Federico Sevilla III : http://jijo.free.net.ph : When we speak of free
Network Administrator : The Leather Collection, Inc. : software we refer to
GnuPG Key ID : 0x93B746BE : freedom, not price.