On Fri, Dec 07, 2001 at 06:41:53PM +0100, Daniel Phillips wrote:
> I've observed disk cache effects with Ext2, the relevant relationship being
> directory entry order vs inode order. Layout of the index itself doesn't
> seem to matter much because of its small size, and 'popularity', which tends
> to keep it in cache.
Exactly.
And if the files have data in them (all my tests were done with files
with bodies) then there is a third data-type (the allocated blocks)
whose order compared to the entry-order and the inode-order also
matters.
> With ReiserFS we see slowdown due to random access even with small
> directories. I don't think this is a cache effect.
I can't see why the benefit from read-ahead on the file-data should be
affected by the directory-size?
I forgot to mention another important effect of hash-ordering:
If you mostly add new files to the directory it is far less work if you
almost always can add the new entry at the end rather than insert it in
the middle. Well, it depends on your implementation of course, but this
effect is quite noticable on reiserfs. When untaring a big directory of
maildir the performance difference between the tea hash and a special
maildir hash was approxemately 20%. The choice of hash should not affect
the performance on writing the data itself, so it has to be related to
the cost of the insert operation.
-- Ragnar Kjørstad Big Storage - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Dec 07 2001 - 21:00:39 EST