It is mostly due to ext2's directory block allocation policy. Files are
allocated close together if they're in the same directory, but
directories are spread more or less randomly over the disk. I have a
program called "treescan" which tries to read in a better order. It's
faster but you still hear a lot of head activity. It's not particularly
random -- it seems even small seeks are noisy.
> I would think that optimizing disk writes would make most of the
> difference go away.
The writes makes some difference: the inodes do seem to get written back
in a bad order. You can see how much improvement is possible here by
mounting with `nodiratime'.
With 2.3.26 the inode write-backs don't seem so significant. (That's
100% subjective and unscientific).
-- Jamie
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/