[Cluster-devel] [GFS2 Patch] pass formal ino in do_filldir_main

[Steven Whitehouse]

Hi,

On Tue, 2007-02-27 at 15:04 -0500, Wendy Cheng wrote:
> Steven Whitehouse wrote:
> > In the mean time no_formal_ino will still exist, its just that it won't
> > be used as the lookup key, so theres no on-disk format change to
> > consider here, all the same fields will continue to have the same
> > values, so there should be no problem to use it again if required at a
> > later date,
> >
> >   
> ok, as long as we don't rush into removing the code, this is an 
> acceptable plan. We can discuss the issue further when we meet 
> face-to-face next month. In the mean time, will go ahead to make lookup 
> code consistent by tentatively putting no_formal_ino aside. Without the 
> changes, half of the current NFS lookups would fail. We need to have 
> something working right now.
> 
Yes, agreed.

> I also invite Kent Baxley (one of our TAMs) here to help out with the 
> issue. I recalled he mentioned sometime ago that the "create" table in 
> GFS2-mySQL benchmark was noticeably slow - not sure whether 
> pick_formal_ino() plays any role there. Will give him two gfs2.ko to 
> re-run the benchmark and see what we get.
> 
> -- Wendy
> 
Probably it doesn't make much difference. I did some tests a while back when trying
to work out why file creates are slow and I discovered that removing the
pick_formal_ino code made no noticeable difference to the create speed,
so the bottleneck appears to be elsewhere,

Steve.