On Mon, 2005-08-15 at 15:06 +0100, Christoph Hellwig wrote:
> On Mon, Aug 15, 2005 at 04:02:03PM +0200, Andi Kleen wrote:
> > Dave Kleikamp <[EMAIL PROTECTED]> writes:
> > > 
> > > jfs no longer uses a directory inode's address space.  Clean up the
> > > code by removing aops for directories altogether
> > 
> > Wouldn't it be better to allow prefetch on directories?
> > After all there is a reason earlykdm does that.

I don't have a suse system in front of me, but doesn't earlykdm preload
files within those directories?  I would think that in most cases, the
directories will be prefetched as a result of preloading those files.

> 
> It's a bit diffcult.  Most filesystems (at least ext3, xfs and jfs)
> don't use the inode's address space for directories but the address_space
> of the underlying block device (or in the xfs case an xfs-private one
> covering the whole block device), so that readahead would have to happen
> on the block device node after finding out where the directory is placed.

exactly.  the sys_readahead interface doesn't lend itself to working
with directories.  Directory inodes would need a dummy set of aops, and
I don't think the benefit would be worth the complexity.
-- 
David Kleikamp
IBM Linux Technology Center



-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf
_______________________________________________
Jfs-discussion mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/jfs-discussion

Reply via email to