[linux-cifs-client] Re: [PATCH 2/3] cifs: use iunique() in cifs_new_inode()

[Jeff Layton]

On Wed, 11 Feb 2009 09:10:35 -0600
Steve French <smfrench at gmail.com> wrote:

> smbfs and ncpfs and fat use iunique but a comment in the kernel source warns
> about using it since it is slow.  Wonder if it is measurably slow?
> 
> On Wed, Feb 11, 2009 at 7:08 AM, Jeff Layton <jlayton at redhat.com> wrote:
> 
> > ...to make sure that we get unique inode numbers when we're generating
> > them on the fly.
> >
> >
> 
> 

I'm sure there will be some performance impact. If you really want to
bench it out correctly, you'll probably need to generate a ton of
inodes on a mount so that you can ensure that each hash bucket will
have a lot. Ideally you'd also want to make the counter in iunique
wrap too so that it has to look for a while for a free inode. 

In practice, I doubt it'll be noticable -- certainly it should still be
faster than a round trip to the server. Even if there is some
performance impact though, I think it's still the right thing to do.

-- 
Jeff Layton <jlayton at redhat.com>