Hi Paul, Ben,

I would certainly not object to having a newer base OS on dowd, although we currently do have some newer compilers installed. There are newer compilers installed in:
  /opt/at7.0/bin/

and you can also use the one in my home directory:
  /home/hfinkel/install/ppc64/gcc/bin

Turning off HT for a few days is fine by me. Please let me know when you're doing it, so that I won't disturb your benchmarks.

 -Hal

----- Original Message -----
> From: "Paul Romano" <paul.k.romano@gmail.com>
> To: "Benjamin S. Allen" <bsallen@alcf.anl.gov>
> Cc: admins@jlse.anl.gov, "Hal J. Finkel" <hfinkel@anl.gov>
> Sent: Tuesday, May 5, 2015 10:20:27 AM
> Subject: Re: [JLSE-Admins] Hardware threading on thing nodes
>
>
> Thanks Ben. RHEL6 or 7 are fine, so whatever is easiest. GCC 4.8.2
> would be sufficient for my needs on dowd.
>
>
> Paul
>
>
> On Tue, May 5, 2015 at 10:05 PM, Allen, Benjamin S. <
> bsallen@alcf.anl.gov > wrote:
>
>
> Hi Paul,
>
> Yea its possible get a thing node with HT disabled. I'll likely grab
> thing03 for this. Hopefully I'll find some time today to change the
> BIOS setting. Do you want RHEL6 or 7 on the box?
>
> We don't have XL licenses unfortunately. If there's a newer version
> on IBM's website I can re-install the eval and get another 90 days
> out of it.
>
> We can also look at installing RHEL7 on the box, which has GCC 4.8.2.
> Hal Finkel (cc'ed) is the primary user of Dowd, so he'll need to
> weigh in on what we should do.
>
> Ben
>
>
>
>
> > On May 4, 2015, at 10:25 PM, Paul Romano < paul.k.romano@gmail.com
> > > wrote:
> >
> > Hi Ben,
> >
> > Just wanted to ping you on this. Do you think it will be possible
> > to get a thing node with HT disabled for a day or two?
> >
> > Also, an unrelated question -- I saw there is an IBM XL compiler on
> > dowd, but it does not have a valid license. Is there another
> > compiler that works on that node? I need something a little more
> > recent than the default gcc 4.4. I've tried unsuccessfully to
> > build a more recent version of gcc.
> >
> > Thanks,
> > Paul
> >
> > On Mon, Apr 27, 2015 at 4:00 PM, Paul Romano <
> > paul.k.romano@gmail.com > wrote:
> > Hi Ben,
> >
> > One aspect of the study I'm performing is to look at any potential
> > differences between the situation you just described, i.e. only
> > using as many threads as there are physical cores with HT enabled,
> > and actually disabling HT altogether, so unfortunately binding to
> > the first hardware thread won't suffice. Once I have everything
> > scripted and ready to go, it shouldn't take more than a day or two
> > to perform the actual runs.
> >
> > Best,
> > Paul
> >
> > On Sun, Apr 26, 2015 at 9:54 PM, Allen, Benjamin S. <
> > bsallen@alcf.anl.gov > wrote:
> > Hi Paul,
> >
> > Will running your application with taskset or numactl suffice, eg:
> > taskset -c 0-35 cmd or numactl -C 0-35 cmd? Thus pinning the
> > application to the first hardware thread on each core.
> >
> > If not how long do you think you'll need a machine with HT
> > disabled?
> >
> > Ben
> >
> > > On Apr 25, 2015, at 7:31 AM, Paul Romano < promano@anl.gov >
> > > wrote:
> > >
> > > Hi,
> > >
> > > I'm currently doing some studies on the impact of hardware
> > > threading on code performance for nuclear reactor neutronic
> > > calculations. I was wondering if it might be possible to have
> > > one of the thing nodes configured with hardware threading
> > > disabled for the sake of running comparisons.
> > >
> > > Thanks,
> > > Paul
> > >
> > > _______________________________________________
> > > Admins mailing list
> > > Admins@lists.jlse.anl.gov
> > > https://lists.jlse.anl.gov/mailman/listinfo/admins
> >
> >
> >
>
>
>

--
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory