Tony Nelson wrote:
At 12:28 AM +0200 10/13/06, Dotan Cohen wrote:
How can I check fragmentation. Googling the subject makes me beleive
that this is not the case in general with Linux.
The common wisdom is that EXT2/3 are not affected by fragmentation, but
without much real-world proof that this is so. The EXT2/3 filesystem
metadata was designed to be not much affected by fragmentation, but that
says little about the file data. I read an article / webpage (that I can't
find right now) by someone who decided to experiment with new and used EXT2
filesystems, and found a substatial slowdown. He was inspired to try this
because he noticed that his computer sped up when given a fresh filesystem.
You could try backing up and restoring to a fresh filesystem. If you
spring for a new computer you'll back up and restore to the new computer.
Either way you'll get a fresh new filesystem.
We've seen significant I/O slowdown on a 4,176-processor Cray XT3 because of
file fragmentation on an ext3 filesystem (overlaid with Lustre). Admittedly the
filesystem in question was about 40 terabytes in size and running under a 2.4
kernel, but fragmentation was there nonetheless. We solved the problem by
rebuilding the filesystem.
See here for a related reference that speaks to improvements in ext3 block
allocation for 2.6.10+ kernels:
http://ext2.sourceforge.net/2005-ols/paper-html/node7.html
