On Jul 23, 2008 08:30 -0500, Jose R. Santos wrote:
> On Tue, 22 Jul 2008 18:58:50 -0600
> Andreas Dilger <[email protected]> wrote:
> > Jose, do you have ext3 results on the same system
> > for the benchmarks you ran? That would tell us how much improvement we
> > get from other ext4 features (e.g. extents vs. block allocation) and how
> > much from flex_bg.
>
> No I dont, I tried doing some runs yesterday but after updating the
> kernel, the results flex_bg are about the same as without it and ext3
> is a lot faster than ext4. Im investigating to see if I messed up the
> kernel build somehow or if we have a regression.
There was another report that the current ext4 code is no longer faster
at compilebench than ext3.
> Valerie did a very comprehensive set of comparisons that could be
> useful for the presentation. I'll try to see if I can recreate this
> once I figure out the regression im seeing but would this work for now?
>
> http://www.bullopensource.org/ext4/20080530/ffsb-readwrite-2.6.26-rc2.html
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
在 2008-07-25五的 18:00 -0400，Andreas Dilger写道：
> On Jul 23, 2008 08:30 -0500, Jose R. Santos wrote:
> > On Tue, 22 Jul 2008 18:58:50 -0600
> > Andreas Dilger <[email protected]> wrote:
> > > Jose, do you have ext3 results on the same system
> > > for the benchmarks you ran? That would tell us how much improvement we
> > > get from other ext4 features (e.g. extents vs. block allocation) and how
> > > much from flex_bg.
> >
> > No I dont, I tried doing some runs yesterday but after updating the
> > kernel, the results flex_bg are about the same as without it and ext3
> > is a lot faster than ext4. Im investigating to see if I messed up the
> > kernel build somehow or if we have a regression.
>
> There was another report that the current ext4 code is no longer faster
> at compilebench than ext3.
>
> > Valerie did a very comprehensive set of comparisons that could be
> > useful for the presentation. I'll try to see if I can recreate this
> > once I figure out the regression im seeing but would this work for now?
> >
> > http://www.bullopensource.org/ext4/20080530/ffsb-readwrite-2.6.26-rc2.html
>
Between 2.6.26-rc2 to now the biggest change is the locking ordering
change (page lock and transaction) and the new ordered mode, and the
delalloc get updated to adopt the new locking order. I wonder if that's
the cause....
Mingming
> Cheers, Andreas
> --
> Andreas Dilger
> Sr. Staff Engineer, Lustre Group
> Sun Microsystems of Canada, Inc.
>