[CIG-LONG] cig-long email list
Eunseo Choi
echoi at ldeo.columbia.edu
Tue May 5 14:46:00 PDT 2009
Garrett Ito wrote:
>
> Hi Eunseo,
>
> Thanks--I'm still working on mine. So if I extrapolate out to say
> 0.1 Myr model time, you'd be at about 1e5yr/(1000*0.1yr)*0.75hrs=750
> hrs (31 days) on six cores? Does SNAC scale well? Can you run the
> job on say 50 cores? I'll probably run it mine on about that number
> of cores.
Hi, Garrett
No hurry. This is not your duty in the first place. I really appreciate
your time and effort.
My previous tests showed that SNAC's efficiency for 50 cores is about 0.6.
So, roughly, 750 hrs*(6/50)/0.6 =150 hrs (6.25 days).
However, the time step size, 0.1 yr, was a highly conservative choice.
It can be safely increased up to about half the Maxwell time, which
would be on the order of 10^8 sec for viscosity = ~1.0e+18 Pa.s and
shear modulus = ~ 10e+10 Pa.
So, when the time step is 1 yr, the estimated computation time on 50
cores would be 0.625 day = 15 hr.
Eunse0
>
> Garrett
>
> ----- Original Message -----
> From: Eunseo Choi <echoi at ldeo.columbia.edu>
> Date: Tuesday, May 5, 2009 4:09 am
> Subject: Re: [CIG-LONG] cig-long email list
> To: Garrett Ito <gito at hawaii.edu>
> Cc: cig-long at geodynamics.org
>
> > Hi, Garrett
> >
> > Here are my results.
> >
> > - 151x45x6 elements.
> > - Run for 1000 time steps with an output frequency of 200.
> > - The time step was constant and 0.1 yr.
> >
> > - Run on 6 cores out of 2 x (Intel Xeon QuadCore CPU E5405 at 2.00GHz).
> > (I don't know the details of the speed of
> > communication between cores).
> > - SNAC was compiled with gcc version 4.1.2 20080704 (Red Hat
> > 4.1.2-44)
> > and optimized with the -O3 option.
> >
> > - Time measured by the 'time' command: 45 min 06 sec.
> >
> > Let me know if any critical piece of information is missing.
> >
> > Eunseo
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > Garrett Ito wrote:
> > >
> > > Hi Enseo and Todd,
> > >
> > > Yes, if detailed comparisons become necessary
> > (i.e., if the two
> > > codes perform within a few factors of eachother) then we
> > should think
> > > about comparing on the same platform. At the moment, I'm
> > having
> > > trouble running my own test case in 2D with the iterative
> > solver I'll
> > > be using for 3D (ugh!). I'll get back to you as
> > soon as I can.
> > >
> > > Garrett
> > >
> > > ----- Original Message -----
> > > From: Eunseo Choi <echoi at ldeo.columbia.edu>
> > > Date: Friday, May 1, 2009 10:42 am
> > > Subject: Re: [CIG-LONG] cig-long email list
> > > To: tehlers at umich.edu
> > > Cc: cig-long-bounces at geodynamics.org, Garrett Ito
> > <gito at hawaii.edu>,
> > > Mark Behn <mbehn at whoi.edu>, cig-long at geodynamics.org
> > >
> > > > Hi, Todd
> > > >
> > > > Thanks for the suggestion.
> > > >
> > > > I've thought about it and think it might be the only proper (not
> > > > just
> > > > better) way of comparing two codes.
> > > > However, I've never succeeded in running Gale so far.
> > > > Also, maybe the comparison is meaningful to some degree if two
> > > > systems
> > > > are not that different?
> > > >
> > > > Eunseo
> > > >
> > > >
> > > >
> > > > Todd Ehlers wrote:
> > > > > Hi Garrett and Eunseo
> > > > > If possible you might want to run both on the same hardware
> > > > and compile both codes with the same compiler with the same
> > > > optimization flags.
> > > > > . Cheers,
> > > > > Todd
> > > > > Sent via BlackBerry from T-Mobile
> > > > >
> > > > > -----Original Message-----
> > > > > From: Eunseo Choi <echoi at ldeo.columbia.edu>
> > > > >
> > > > > Date: Fri, 01 May 2009 08:39:51
> > > > > To: Garrett Ito<gito at hawaii.edu>
> > > > > Cc: Mark Behn<mbehn at whoi.edu>; <cig-long at geodynamics.org>
> > > > > Subject: Re: [CIG-LONG] cig-long email list
> > > > >
> > > > >
> > > > > Hi, Garrett
> > > > >
> > > > > I agree that our first order goal should be to compare
> > performance.> > > My model will have 150x45x6 elements and be
> > made as similar to
> > > > yours as
> > > > > possible in terms of initial/boundary conditions.
> > > > > I'll get back to you when I get results.
> > > > > Thanks.
> > > > >
> > > > > Eunseo
> > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > Garrett Ito wrote:
> > > > >
> > > > >> Hi Enseo,
> > > > >>
> > > > >> Regarding your question about viscosity--
> > > > remember that Gale is a
> > > > >> purely viscous code so it simulates plasticity with variable,
> > > > >> strain-dependent viscosity (i.e.,
> > pseudoplasticity). As
> > > > for the
> > > > >> benchmark, at this point I'm most interested in seeing
> > > > whether the
> > > > >> two codes are comparable in speed (or shall I say
> > > > "slowness"). What
> > > > >> will be most important at this point is that the plasticity
> > > > law uses
> > > > >> generates localized zones of strains like faults. Yes,
> > > > I can produce
> > > > >> a whole suit of 3D calculations with whatever number of
> > > > elements in
> > > > >> the 3rd dimension, plus details on cpu #'s and compute
> > > > times. I'll
> > > > >> start one soon with 150x45x6 (i.e., I think Gale will
> > want at
> > > > least 4
> > > > >> elements in the 3rd dimension). Lemme know how your
> > > > side is coming
> > > > >> and we can work on details of the comparisons.
> > > > >>
> > > > >> Thanks,
> > > > >>
> > > > >> Garrett
> > > > >>
> > > > >> Eunseo Choi wrote:
> > > > >>
> > > > >>> Hi, Garrett
> > > > >>>
> > > > >>> First of all, thanks for responding to my comments and
> > > > suggesting an
> > > > >>> interesting benchmark.
> > > > >>>
> > > > >>> SNAC cannot currently handle Drucker-Prager plasticity, so
> > > > I'll use
> > > > >>> Mohr-Coulomb model.
> > > > >>>
> > > > >>> A more fundamental issue is that SNAC is hardwired to be 3D,
> > > > which means
> > > > >>> it can solve a 2D problem using a thin 3D model with plane
> > > > strain
> > > > >>> conditions.
> > > > >>> A pseudo-2D model like this will obviously raise the
> > > > numerical cost
> > > > >>> compared to a true 2D one, rendering the benchmark unfair.
> > > > >>>
> > > > >>> So, could you please set up a one-element thick 3D model
> > > > from your 2D model?
> > > > >>> Let me know if it's too much work on your side.
> > > > >>>
> > > > >>> I also need to know how long in terms of model time you ran
> > > > this model,
> > > > >>> if it was run on a single processor, and how many time steps
> > > > were taken.
> > > > >>>
> > > > >>> Finally, here is a non-technical (I guess) question.
> > > > >>> According to the Gale's manual, the Frank Kamenetskii
> > > > viscosity is
> > > > >>> defined as temperature-dependent only.
> > > > >>> Then, why do we see the viscosity field varying consistently
> > > > with the
> > > > >>> strain rate field?
> > > > >>> Such a correspondence is expected when viscosity is also
> > > > stress-dependent.
> > > > >>> Am I misunderstanding something?
> > > > >>>
> > > > >>>
> > > > >>> Best,
> > > > >>> Eunseo
> > > > >>>
> > > > >>>
> > > > >>>
> > > > >>> Garrett Ito wrote:
> > > > >>>
> > > > >>>
> > > > >>>> Hi Eunseo,
> > > > >>>>
> > > > >>>> Aha! my little comment did catch your eye--
> > > > good. Your suggestion of
> > > > >>>> comparing Gale and SNAC is a great one and I'd like to
> > > > try. Here's
> > > > >>>> one we could look at. Its a simple case rifting in
> > > > which extension is
> > > > >>>> driven by the two side boundaries apart, each at ~12.6
> > > > km/Myr (so full
> > > > >>>> spreading rate is 25.2 km/Myr). Free inflow at the
> > > > base and top free
> > > > >>>> to move up and down. Frank Kamenetskii
> > temperature-
> > > > dependent
> > > > >>>> viscosity with Drucker Prager yielding. The
> > > > temperature structure has
> > > > >>>> T=273K at the surface and 1473K at the base with a linear
> > > > profile in
> > > > >>>> between. A box of postfailure strain is imposed at
> > > > t=0 to control
> > > > >>>> where the faults initiate. The attached pdf shows
> > > > Gale's solution in
> > > > >>>> 2D with 150x45 elements. I also attached Gale's input
> > > > file so you can
> > > > >>>> see the rheology parameters. How does this sound?
> > > > >>>>
> > > > >>>> Garrett
> > > > >>>>
> > > > >>>> ----- Original Message -----
> > > > >>>> From: Eunseo Choi <echoi at ldeo.columbia.edu>
> > > > >>>> Date: Wednesday, April 29, 2009 5:34 pm
> > > > >>>> Subject: Re: [CIG-LONG] cig-long email list
> > > > >>>> To: cig-long at geodynamics.org
> > > > >>>>
> > > > >>>>
> > > > >>>>
> > > > >>>>> Hi, Garrett
> > > > >>>>>
> > > > >>>>> SNAC is a 3D code like Gale so it is supposed to be
> > > > >>>>> computationally
> > > > >>>>> expensive compared to any 2D codes.
> > > > >>>>> I believe this is not what you meant.
> > > > >>>>>
> > > > >>>>> As far as I know, the performance of SNAC and Gale has never
> > > > >>>>> been
> > > > >>>>> compared quantitatively yet. Please let me know if
> > it's done
> > > > >>>>> even
> > > > >>>>> qualitatively.
> > > > >>>>> So, I wonder why you have the impression that SNAC is too
> > > > >>>>> expensive for
> > > > >>>>> 3D models.
> > > > >>>>>
> > > > >>>>> By the way, SNAC can do Mohr-Coulomb plasticity in 3D and
> > > > >>>>> doesn't have
> > > > >>>>> the restriction that plastic flow should always be
> > > > incompressible.>>>>>
> > > > >>>>> Eunseo
> > > > >>>>>
> > > > >>>>>
> > > > >>>>> Garrett Ito wrote:
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>> Hi Mark,
> > > > >>>>>>
> > > > >>>>>> Good-- I'm glad to see 3D is starting to
> > get a
> > > > >>>>>>
> > > >
> > >>>>>>
> > > > >>>>> footing in the
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>> community. This is my main reason for using
> > Gale as I
> > > > >>>>>>
> > > >
> > >>>>>>
> > > > >>>>> already do
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>> faulting in 2D with FLAC (and it is my understanding
> > that SNAC
> > > > >>>>>>
> > > >
> > >>>>>>
> > > > >>>>> is too
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>> computationally expensive in 3D ?). Correction...
> > > > Mohr-
> > > > >>>>>>
> > > >
> > >>>>>>
> > > > >>>>> Coulomb has
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>> been removed from Gale because it was unstable in 3D
> > and is
> > > > >>>>>>
> > > >
> > >>>>>>
> > > > >>>>> identical
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>> to Drucker-Prager in 2D.
> > > > >>>>>>
> > > > >>>>>> Garrett
> > > > >>>>>>
> > > > >>>>>> ----- Original Message -----
> > > > >>>>>> From: Mark Fleharty <mark.fleharty at gmail.com>
> > > > >>>>>> Date: Wednesday, April 29, 2009 12:46 pm
> > > > >>>>>> Subject: Re: [CIG-LONG] cig-long email list
> > > > >>>>>> To: Garrett Ito <gito at hawaii.edu>
> > > > >>>>>> Cc: cig-long at geodynamics.org
> > > > >>>>>>
> > > > >>>>>>
> > > >
> > >>>>>>
> > > > >>>>>>> Garrett,
> > > > >>>>>>>
> > > > >>>>>>> It is good to hear from someone else who is doing 3D
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>> simulations in
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>>> Gale. I am also doing 3D simulations with Gale.
> > > > >>>>>>> Welcome aboard.
> > > > >>>>>>>
> > > > >>>>>>> I thought that Mohr-Coulomb doesn't work in Gale in 3D.
> > > > >>>>>>> Was that
> > > > >>>>>>> something that got fixed?
> > > > >>>>>>>
> > > > >>>>>>> Mark
> > > > >>>>>>>
> > > > >>>>>>> On Wed, Apr 29, 2009 at 4:27 PM, Garrett Ito
> > > > >>>>>>> <gito at hawaii.edu> wrote:
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> Dear colleagues,
> > > > >>>>>>>>
> > > > >>>>>>>> I would like to introduce myself as a
> > new to
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>> the cig-long
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>>> list and new
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> GALE user. I would like to encourage us all
> > to use
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>> the cig-
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>>> long list more
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> often to help spread the knowledge of the CIG codes.
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>> I think
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>>> it would be
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> helpful, for example, we made it a practice of sending
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> individual questions
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> to cig-long along with questions you send to
> > Walter/Eun-seo
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> about the
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> various coding issues.
> > > > >>>>>>>>
> > > > >>>>>>>> Just to start a conversation...I have started using
> > > > GALE--
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> really in
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> collaboration with Walter Landry--over the last half
> > > > year with
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> interests in
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> applications of studying faulting and magmatism at
> > mid-ocean
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> ridges in 3D.
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> A lot of my work to date has been to experimenting
> > with the
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> Mohr-Coulomb,
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> Drucker-Prager, and StrainWeakening (e.g., damage
> > laws and
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> healing) and
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> understanding the faulting behavior predicted by Gale.
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> Recently, we have
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> started running jobs in 3D, which I have found to be
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> surprisingly cpu
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> intensive. I view improving the performance
> > of Gale
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>> for 3D faulting
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>>>> problems as an important goal and on which I will
> > be working
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> closely with
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>>>> Walter and the Underworld group over the next year.
> > > > >>>>>>>>
> > > > >>>>>>>> I look forward to hearing more from you.
> > > > >>>>>>>>
> > > > >>>>>>>> Garrett
> > > > >>>>>>>> --
> > > > >>>>>>>>
> > > > >>>>>>>> =====================================
> > > > >>>>>>>>
> > > > >>>>>>>> Prof. Garrett Ito
> > > > >>>>>>>>
> > > > >>>>>>>> SOEST, Geology & Geophysics
> > > > >>>>>>>>
> > > > >>>>>>>> 1680 East-West Rd., POST 810
> > > > >>>>>>>>
> > > > >>>>>>>> Univ. of Hawaii
> > > > >>>>>>>>
> > > > >>>>>>>> Honolulu, HI 96822 USA
> > > > >>>>>>>>
> > > > >>>>>>>>
> > > > >>>>>>>>
> > > > >>>>>>>> 808-956-9717
> > > > >>>>>>>>
> > > > >>>>>>>> gito at hawaii.edu
> > > > >>>>>>>>
> > > > >>>>>>>> www.soest.hawaii.edu/GG/FACULTY/ITO
> > > > >>>>>>>>
> > > > >>>>>>>> ======================================
> > > > >>>>>>>>
> > > > >>>>>>>> _______________________________________________
> > > > >>>>>>>> CIG-LONG mailing list
> > > > >>>>>>>> CIG-LONG at geodynamics.org
> > > > >>>>>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
> > > > >>>>>>>>
> > > > >>>>>>>>
> > > >
> > >>>>>>>>
> > > >
> > >>>>>>>>
> > > > >>>>>>> --
> > > > >>>>>>> Mark Fleharty
> > > > >>>>>>> +1 505 217-2089
> > > >
> > >>>>>>>
> > > >
> > >>>>>>>
> > > > >>>>>> ------------------------------------------------------
> > ----
> > > > -----
> > > > >>>>>>
> > > >
> > >>>>>>
> > > > >>>>> ---------
> > > > >>>>>
> > > > >>>>>
> > > > >>>>>> _______________________________________________
> > > > >>>>>> CIG-LONG mailing list
> > > > >>>>>> CIG-LONG at geodynamics.org
> > > > >>>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
> > > > >>>>>>
> > > > >>>>>>
> > > >
> > >>>>>>
> > > > >>>>> --
> > > > >>>>> =====================================
> > > > >>>>> Eunseo Choi
> > > > >>>>> Post-Doctoral Research Scientist
> > > > >>>>>
> > > > >>>>> Lamont-Doherty Earth Observatory
> > > > >>>>> 303f Oceanography
> > > > >>>>> P.O. Box 1000
> > > > >>>>> 61 Rt. 9W
> > > > >>>>> Palisades, NY 10964-8000
> > > > >>>>>
> > > > >>>>> http://www.ldeo.columbia.edu/~echoi
> > > > >>>>> Phone: 845-365-8813 Fax: 845-365-8156
> > > > >>>>> =====================================
> > > > >>>>>
> > > > >>>>> _______________________________________________
> > > > >>>>> CIG-LONG mailing list
> > > > >>>>> CIG-LONG at geodynamics.org
> > > > >>>>> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-long
> > > > >>>>>
> > > > >>>>>
> > > > >>>
> > > > >>>
> > > > >> --
> > > > >>
> > > > >> =====================================
> > > > >>
> > > > >> Prof. Garrett Ito
> > > > >>
> > > > >> SOEST, Geology & Geophysics
> > > > >>
> > > > >> 1680 East-West Rd., POST 810
> > > > >>
> > > > >> Univ. of Hawaii
> > > > >>
> > > > >> Honolulu, HI 96822 USA
> > > > >>
> > > > >>
> > > > >>
> > > > >> 808-956-9717
> > > > >>
> > > > >> gito at hawaii.edu
> > > > >>
> > > > >> www.soest.hawaii.edu/GG/FACULTY/ITO
> > > > >>
> > > > >> ======================================
> > > > >>
> > > > >>
> > > > >
> > > > >
> > > > >
> > > >
> > > >
> > > > --
> > > > =====================================
> > > > Eunseo Choi
> > > > Post-Doctoral Research Scientist
> > > >
> > > > Lamont-Doherty Earth Observatory
> > > > 303f Oceanography
> > > > P.O. Box 1000
> > > > 61 Rt. 9W
> > > > Palisades, NY 10964-8000
> > > >
> > > > http://www.ldeo.columbia.edu/~echoi
> > > > Phone: 845-365-8813 Fax: 845-365-8156
> > > > =====================================
> > > >
> >
> >
> > --
> > =====================================
> > Eunseo Choi
> > Post-Doctoral Research Scientist
> >
> > Lamont-Doherty Earth Observatory
> > 303f Oceanography
> > P.O. Box 1000
> > 61 Rt. 9W
> > Palisades, NY 10964-8000
> >
> > http://www.ldeo.columbia.edu/~echoi
> > Phone: 845-365-8813 Fax: 845-365-8156
> > =====================================
> >
--
=====================================
Eunseo Choi
Post-Doctoral Research Scientist
Lamont-Doherty Earth Observatory
303f Oceanography
P.O. Box 1000
61 Rt. 9W
Palisades, NY 10964-8000
http://www.ldeo.columbia.edu/~echoi
Phone: 845-365-8813 Fax: 845-365-8156
=====================================
More information about the CIG-LONG
mailing list