[Cluster-devel] Re: Why the gfs2 performance regressed?

[Steven Whitehouse]

Hi,

On Thu, 2008-01-03 at 00:44 +0800, Cheng Renquan wrote:
> On Jan 2, 2008 6:19 PM, Steven Whitehouse <swhiteho at redhat.com> wrote:
> > Are you running single node? if so then use lock_nolock rather than
> > lock_dlm as it will be much faster for fcntl locks. Even if you intend
> > to run as a cluster eventually, a single node comparison against
> > lock_nolock would be useful to try and eliminate some possibilities,
> No, I'm using a two nodes environment, the /mnt/gfs2 are gfs2 mounting
> points on both two nodes, and they are both samba shared folders.
> And in fact, the two nodes are using a clustered LVM from the same
> iSCSI device, so dlm would be the only lock manager?
> 
> Actually What I cannot explain is merely that the only difference with
> the kernel: from 2.6.18-53.el5 (stocked with RHEL51) to latest
> gfs2-nmw.git, I don't know why.
> 

There was a performance regression with the -nmw tree relating to a set
of patches which I removed from the tree this morning. That would make
it slower, but not by the amount that you are seeing, and also it would
affect only "normal" I/O and not fcntl locks.

Is there a point in the history of the  -nmw git tree which you know is
ok? Perhaps it would be possible to bisect the tree to find the problem
patch?

Steve.