EXT3 performance on Large (multi-TeraByte) RAID

Andreas Dilger adilger at clusterfs.com
Thu May 13 06:08:33 UTC 2004

On May 12, 2004  19:09 -0700, M K wrote:
> Has anyone experienced a significant degradation in ext3 performance
> when using it on a Multi-TeraByte RAID? As part of an experimental setup,
> I hooked up three 300GB drives and made an EXT3 RAID5 out of them, using
> the entire space one each drive, and started throwing a large number
> of files in the size-range 3KB to 50 KB. Then, I deleted the raid, and
> created a new one, but this time, I used only 3 Gigs from each drive (a
> very small RAID compared to the earlier one). After repeating the same
> test, a huge improvement in performance was see - hence, the question:
> does ext3 performance degrade significantly as the file system size
> increases?

Are you using 2.4 or 2.6 kernels?  In the 2.4 kernel files are allocated
evenly across all of the filesystem space.  However, for large filesystems
this is not very effective.  In 2.6 kernels the Orlov allocator will keep
files created by the same process (e.g. tar) in a single group if possible,
which localizes block allocation and avoids seeks.

Cheers, Andreas
Andreas Dilger

More information about the Ext3-users mailing list