[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: htree stabilitity and performance issues



On Fri, Dec 19, 2003 at 05:29:18PM +0000, Stephen C. Tweedie wrote:
> Hi,
> 
> On Fri, 2003-12-19 at 16:56, Andreas Dilger wrote:
> 
> > Doesn't readdir order == hash order with Ted's NFS fixes for htree?  In
> > that case, if you copied the files in readdir order to a new directory
> > the inodes would also be in hash order on disk, until you add a new file.
> > It might be worthwhile for a large directory that changes very little,
> > but probably a waste of time otherwise.
> 
> I'm so used to tools sorting readdir output themselves, I'd forgotten
> that cp would leave the order intact, but I just checked and yes, it
> does.  So yes, that should work, unless of course you're using Ted's
> preload to resort readdir!

The preload hack should still work, because we're sorting on inode
number, which will result in an optimal order to access the files.
The nice thing about the preload hack is that it always should help
--- even in some cases where htree isn't in use (for example, in a
large maildir directory after a long period of time, the order of the
directory entries and the order of disk inodes will become more and
more randomized, so using the preload hack should help in that case as
well).

						- Ted





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]