[PVFS-users] version 1.6.2 slower than 1.60?

[Jeremy Enos]

I've finally had the time to test... and I'm seeing that 1.6.1 is 
performing as poorly as 1.6.2.  Both are terrible drops relative to 1.6.0, 
with the crude dd test I've described at the start of this thread.
I'd really like to get something newer into OSCAR than 1.6.0, but with this 
performance problem, I'm inclined to leave it at 1.6.0 for the time 
being.  What should I try next?
thx-

         Jeremy

At 11:13 AM 1/15/2004, Rob Ross wrote:
>Hi Jeremy,
>
>If you have time, could you try 1.6.1?  We've talked with some other
>users, and we haven't heard any other reports of this...
>
>Rob
>
>On Tue, 13 Jan 2004, Jeremy Enos wrote:
>
> > Nothing else changed on the system.  I've put 1.6.0 back on again and
> > verified an improvement.  I use the same configure options for each
> > one.  Snipped from my spec file:
> >
> > %install
> > # Make and install PVFS
> > cd $RPM_BUILD_DIR/%{name}-%{version}
> > ./configure --prefix=%{buildroot}/%{prefix} --enable-lfs
> >
> > # Make and install PVFS kernel module
> > ./configure --prefix=%{buildroot}/%{prefix} \
> >    --with-pvfs=%{buildroot}/%{prefix} \
> >    --with-moddir=%{buildroot}/lib/modules/%{kernel}/kernel/fs/pvfs \
> >    --with-kernel-headers=%{kernel_source}/include \
> >    --with-newstyle \
> >    --sbindir=%{buildroot}/%{prefix}/sbin \
> >
> > Let me know if you need anything else.  Thx-
> >
> >          Jeremy
> >
> > At 12:02 PM 1/13/2004, Rob Ross wrote:
> > >Hi Jeremy,
> > >
> > >What else changed in the system?  What configure options did you use?
> > >Have you since re-run the 1.6.0 numbers and get the same thing?
> > >
> > >Thanks for the report,
> > >
> > >Rob
> > >
> > >On Tue, 13 Jan 2004, Jeremy Enos wrote:
> > >
> > > > Hi folks-
> > > > I did crude benchmark comparisons using dd to write to a filesystem
> > > mounted
> > > > via the pvfsd kernel module.  I had just moved from version 1.6.0 to
> > > 1.6.2,
> > > > and the performance dropped over 90%.  Could I be doing something 
> wrong?
> > > > Stat of test:
> > > > # of iod: 16
> > > > # of clients: 64
> > > > block size: 65536
> > > > count: 2000
> > > >
> > > > I did concurrent writes/reads and timed from start to finish, and then
> > > > calculate an aggregate dtr.  I know it's a crude test, but it's a
> > > > repeatable difference between 1.6.0 and 1.6.2.
> > > > Let me know if you have any suggestions.
> > > >
> > > >       Jeremy
> > >
> > >_______________________________________________
> > >PVFS-users mailing list
> > >PVFS-users at www.beowulf-underground.org
> > >http://beowulf-underground.org/mailman/listinfo/pvfs-users
> >
> >