Messages in this thread | | | Subject | Re: [Ext2-devel] Re: [Bug 417] New: htree much slower than regular ext3 | From | "Stephen C. Tweedie" <> | Date | 13 Mar 2003 21:04:15 +0000 |
| |
Hi,
On Thu, 2003-02-27 at 21:00, Andreas Dilger wrote:
> I've got a patch which should help here, although it was originally written > to speed up the "create" case instead of the "lookup" case. In the lookup > case, it will do a pre-read of a number of inode table blocks, since the cost > of doing a 64kB read and doing a 4kB read is basically the same - the cost > of the seek.
No it's not --- you're evicting 16 times as much other potentially-useful data from the cache for each lookup. You'll improve the "du" or "ls -l" case by prefetching, but you may well slow down the overall system performance when you're just doing random accesses (eg. managing large spools.)
It would be interesting to think about how we can spot the cases where the prefetch is likely to be beneficial, for example by observing "stat"s coming in in strict hash order.
--Stephen
- 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/
| |