On 2/8/06, David Guerizec <[EMAIL PROTECTED]> wrote:
> So the directories sensible to fragmentation are IMHO, /var/tmp
> and /usr/portage, and they are the ones to put on different partitions.

I tend to agree with you.  They should also be no larger than
necessary.  My current partition setup is:

/boot                100M
/                      6G
/tmp                   2G
/var                   5G
/usr/portage           1G
/usr/portage/packages  6G (includes distfiles)
/usr/src               2G
/home                 66G (the rest of the disk)

My feeling is that if you partition sensibly, you don't need to worry
much about fragmentation.

> Now, I don't have exact numbers to prove my sayings, but anyone can make the
> test themself, if they already have /var/tmp and/or /usr/portage on separate
> partitions.

_Many_ people have reported speedups from having portage on a small
separate volume/partition.

> I didn't have time yet to sort out what kind of filesystem is more or less
> sensible to fragmentation, but from my experience, ext[23] is not a good
> candidate for /var/tmp or /usr/postage. Reiser3 has proven to fragment too,
> and one of the last system I installed was formated with XFS, which I will
> "defragment" in a few weeks. Hopefully I could then come with numbers.

XFS is probably the most resistant to fragmentation due to it's
extents based format and delayed allocation feature.  But in all of
the tests I've done on my system, xfs was slower than ext3 for the
tasks I care about (backup and restore performance).  But there were
some improvements in xfs in 2.6.16 that I have been meaning to try
out.

There is an excellent paper written by the ext3 authors on what they
are working on.  It gives a good basis for understanding the tradeoffs
between xfs, ext3, and reiserfs and their different features:

http://ext2.sourceforge.net/2005-ols/paper-html/index.html

-Richard

-- 
[email protected] mailing list

Reply via email to