[DFTB-Plus-User] DFTB-Plus-User Digest, Vol 58, Issue 11

Ben Hourahine benjamin.hourahine at strath.ac.uk
Sat Jun 22 21:25:50 CEST 2019


Hello David,

If possible use an optimized blas instead of the reference one, as this
makes a big difference in performance.

You might be better off using a serial compiled binary and 2 threads on
each of the 4 images (if with a thread aware blas/lapack).

Yes, sorry, had picked up 8 cores and mistakenly thought that was the
number of images (is 4 enough for REMD?).

Regards

Ben

On 22/06/2019 17:34, David Furman wrote:
> Hi Ben,
>
> 1. I use libblas-dev 3.6.0 (This  implementation is the Fortran 77
> reference implementation found  at netlib)
> 2. I run each instance from the same shell, 'echo $OMP_NUM_THREADS'
> yields: 1
> 3. Yes it is the correct context. I try to run a Replica-Exchange MD
> using 4 replicas. As far as I know, i-pi distributes the replicas to
> clients (DFTB+ in this case) so that I can run 4 concurrent replicas.
> Since I have a relatively large system in terms of number of atoms,
> each replica takes a significant amount of time. Hence I was thinking
> of parallelizing each replica on 2 or more MPI cores (i.e. parallelize
> each DFTB+ instance). But, then I encounter this weird "bottleneck".
> Regarding your suggestion: I'm not sure how running 8 serial instances
> is of help in my case.
>
> I appreciate your help!
>
>  
>
> On Sat, Jun 22, 2019 at 2:54 PM
> <dftb-plus-user-request at mailman.zfn.uni-bremen.de
> <mailto:dftb-plus-user-request at mailman.zfn.uni-bremen.de>> wrote:
>
>     Send DFTB-Plus-User mailing list submissions to
>             dftb-plus-user at mailman.zfn.uni-bremen.de
>     <mailto:dftb-plus-user at mailman.zfn.uni-bremen.de>
>
>     To subscribe or unsubscribe via the World Wide Web, visit
>            
>     https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
>
>     or, via email, send a message with subject or body 'help' to
>             dftb-plus-user-request at mailman.zfn.uni-bremen.de
>     <mailto:dftb-plus-user-request at mailman.zfn.uni-bremen.de>
>
>     You can reach the person managing the list at
>             dftb-plus-user-owner at mailman.zfn.uni-bremen.de
>     <mailto:dftb-plus-user-owner at mailman.zfn.uni-bremen.de>
>
>     When replying, please edit your Subject line so it is more specific
>     than "Re: Contents of DFTB-Plus-User digest..."
>
>
>     Today's Topics:
>
>        1. Possible bug in DFTB+ with MPI (David Furman)
>        2. Re: problem with installation (Farzad Molani)
>        3. Re: Possible bug in DFTB+ with MPI (Ben Hourahine)
>
>
>     ----------------------------------------------------------------------
>
>     Message: 1
>     Date: Sat, 22 Jun 2019 14:41:01 +0100
>     From: David Furman <sirok4 at gmail.com <mailto:sirok4 at gmail.com>>
>     To: dftb-plus-user at mailman.zfn.uni-bremen.de
>     <mailto:dftb-plus-user at mailman.zfn.uni-bremen.de>
>     Subject: [DFTB-Plus-User] Possible bug in DFTB+ with MPI
>     Message-ID:
>            
>     <CADtBuo94A729tJFsTW5ev69sEs4qVNM=qOXkjMKFTS8_f1N8pg at mail.gmail.com
>     <mailto:qOXkjMKFTS8_f1N8pg at mail.gmail.com>>
>     Content-Type: text/plain; charset="utf-8"
>
>      Hi all,
>     I have a question concerning the running of DFTB+ with MPI. There
>     seems to
>     be a problem when running more than one DFTB+ instance (with MPI
>     parallelization) on the same system.
>
>     When I run 4 instances (separate folders) of dftb+ with 2 cores
>     each (i.e.
>     8 cores in total):
>     mpirun.openmpi -np 2 dftb+, each %CPU usage drops to 50%.
>
>     This is the 'top' output:
>     ========================================================
>      35426 user    20   0  423384  40404  18248 R  50.5  0.0   0:37.69
>     dftb+
>      35434 user    20   0  423080  39720  18248 R  50.2  0.0   0:05.70
>     dftb+
>      35396 user    20   0  422044  38660  17644 R  49.8  0.0   1:21.35
>     dftb+
>      35402 user    20   0  423384  40460  18304 R  49.8  0.0   1:09.54
>     dftb+
>      35403 user    20   0  422044  38872  17848 R  49.8  0.0   1:16.15
>     dftb+
>      35427 user    20   0  422044  39048  18032 R  49.8  0.0   0:32.93
>     dftb+
>      35435 user    20   0  421732  38164  17716 R  49.8  0.0   0:09.07
>     dftb+
>      35395 user    20   0  423384  40360  18204 R  49.5  0.0   1:13.18
>     dftb+
>     =========================================================
>
>     Whereas, when I run one instance with 8 cores, the efficiency is
>     100% as
>     expected:
>
>     =========================================================
>      35837 user    20   0  441868  35196  19112 R 100.3  0.0   0:04.90
>     dftb+
>      35828 user    20   0  442580  35676  19120 R 100.0  0.0   0:04.83
>     dftb+
>      35829 user    20   0  442544  36024  19432 R 100.0  0.0   0:04.91
>     dftb+
>      35830 user    20   0  441912  35356  19360 R 100.0  0.0   0:04.91
>     dftb+
>      35831 user    20   0  441688  35048  19304 R 100.0  0.0   0:04.90
>     dftb+
>      35833 user    20   0  441904  35656  19500 R 100.0  0.0   0:04.91
>     dftb+
>      35841 user    20   0  441516  34824  19076 R 100.0  0.0   0:04.89
>     dftb+
>      35843 user    20   0  441224  34456  18868 R  99.7  0.0   0:04.89
>     dftb+
>     =========================================================
>
>     I run both cases with OMP_NUM_THREADS=1.
>     with open-mpi 1.10.2 and gcc 5.4.0.
>
>     Could anyone give a hint about what is wrong?
>     -------------- next part --------------
>     An HTML attachment was scrubbed...
>     URL:
>     <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20190622/96b0d502/attachment-0001.html>
>
>     ------------------------------
>
>     Message: 2
>     Date: Sat, 22 Jun 2019 21:42:46 +0800
>     From: Farzad Molani <f.molani at gmail.com <mailto:f.molani at gmail.com>>
>     To: "User list for DFTB+ related questions"
>             <dftb-plus-user at mailman.zfn.uni-bremen.de
>     <mailto:dftb-plus-user at mailman.zfn.uni-bremen.de>>
>     Subject: Re: [DFTB-Plus-User] problem with installation
>     Message-ID:
>            
>     <CAJY0Q+obk-e8iUOJ_986DS4tWpWG4kAZ99a=6qNstswS0myifg at mail.gmail.com
>     <mailto:6qNstswS0myifg at mail.gmail.com>>
>     Content-Type: text/plain; charset="utf-8"
>
>     I know,  but I don't know how can I unstall arpack?
>     The package downloaded and now I need help to install it.
>     Please note that, I don't permit to login as a root.
>
>     On Saturday, June 22, 2019, RICCARDO ROZZA
>     <rozza.riccardo at studium.unict.it
>     <mailto:rozza.riccardo at studium.unict.it>>
>     wrote:
>
>     > Dear Farzad,
>     >
>     > It seems that you need to install the arpack library (ld: cannot
>     find
>     > -larpack).
>     > All the best,
>     >
>     > Riccardo.
>     >
>     > On 22 Jun 2019, at 10:01, Farzad Molani <f.molani at gmail.com
>     <mailto:f.molani at gmail.com>> wrote:
>     >
>     >
>     > Dear DFTB users,
>     > I want to install the newest version of DFTB+. The source and
>     executable
>     > files from http://www.dftbplus.org/download/dftb-stable/  have been
>     > downloaded and untarred. I try to install the software based on
>     > instructions on INSTALL.rst, but I got the following error after
>     the make.
>     > .
>     > .
>     > .
>     > .
>     > w.o dispuffdata.o inputdata.o pmlocalisation.o repcont.o repspline.o
>     > reppoly.o xlbomd.o extlagrangian.o linresp.o linrespgrad.o qm.o
>     > linrespcommon.o arpack.o eigensolver.o typegeometry.o simplemixer.o
>     > numderivs2.o potentials.o thirdorder.o charges.o dftbplusu.o
>     geoopt.o
>     > steepdesc.o gdiis.o orbitalequiv.o broydenmixer.o stress.o etemp.o
>     > hermite.o factorial.o intrinsicpr.o mixer.o andersonmixer.o
>     > elecconstraints.o angmomentum.o eigenvects.o parser.o hsdutils2.o
>     > unitconversion.o oldcompat.o typegeometryhsd.o oldskdata.o
>     densitymatrix.o
>     > populations.o repulsive.o spinorbit.o forces.o 
>     -L/home/fmolani/DFTB+/
>     > dftbplus-18.2/_build/external/xmlf90 -lxmlf90 
>     -L/home/fmolani/DFTB+/
>     > dftbplus-18.2/_build/external/fsockets -lfsockets
>     > -L/apps/intel/2015/composer_xe_2015.3.187/mkl/lib/intel64
>     > -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core  -larpack -liomp5
>     > ld: cannot find -larpack
>     > make[1]: *** [dftb+] Error 1
>     > make[1]: Leaving directory `/home/fmolani/DFTB+/dftbplus-
>     > 18.2/_build/prog/dftb+'
>     > make: *** [dftb+] Error 2
>     >
>     > I would appreciate to help me to solve the problem.
>     > Best regards, Farzad
>     > --
>     > Farzad Molani
>     > Postdoctoral Research Fellow
>     > School of Pharmaceutical Science and Technology
>     > Tianjin University
>     > Building 24, Room A205
>     > 92 Weijin Road, Nankai District
>     >
>     <https://www.google.com/maps/search/92+Weijin+Road,+Nankai+District+%0D%0A+P.+R.+China?entry=gmail&source=g>
>     > P. R. China
>     >
>     <https://www.google.com/maps/search/92+Weijin+Road,+Nankai+District+%0D%0A+P.+R.+China?entry=gmail&source=g>
>     > Tel.: +8617526526420
>     > Email: f.molani at gmail.com <mailto:f.molani at gmail.com> and
>     f.molani at tju.edu.cn <mailto:f.molani at tju.edu.cn>
>     > <fmolani at dena.kntu.ac.ir <mailto:fmolani at dena.kntu.ac.ir>>
>     >
>     >
>     > --
>     > Farzad Molani
>     > Postdoctoral Research Fellow
>     > School of Pharmaceutical Science and Technology
>     > Tianjin University
>     > Building 24, Room A205
>     > 92 Weijin Road, Nankai District
>     >
>     <https://www.google.com/maps/search/92+Weijin+Road,+Nankai+District+%0D%0A+P.+R.+China?entry=gmail&source=g>
>     > P. R. China
>     >
>     <https://www.google.com/maps/search/92+Weijin+Road,+Nankai+District+%0D%0A+P.+R.+China?entry=gmail&source=g>
>     > Tel.: +8617526526420
>     > Email: f.molani at gmail.com <mailto:f.molani at gmail.com> and
>     f.molani at tju.edu.cn <mailto:f.molani at tju.edu.cn>
>     > <fmolani at dena.kntu.ac.ir <mailto:fmolani at dena.kntu.ac.ir>>
>     > _______________________________________________
>     > DFTB-Plus-User mailing list
>     > DFTB-Plus-User at mailman.zfn.uni-bremen.de
>     <mailto:DFTB-Plus-User at mailman.zfn.uni-bremen.de>
>     >
>     https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
>     >
>     >
>     >
>
>     -- 
>     Farzad Molani
>     Postdoctoral Research Fellow
>     School of Pharmaceutical Science and Technology
>     Tianjin University
>     Building 24, Room A205
>     92 Weijin Road, Nankai District
>     P. R. China
>     Tel.: +8617526526420
>     Email: f.molani at gmail.com <mailto:f.molani at gmail.com> and
>     f.molani at tju.edu.cn <mailto:f.molani at tju.edu.cn>
>     <fmolani at dena.kntu.ac.ir <mailto:fmolani at dena.kntu.ac.ir>>
>     -------------- next part --------------
>     An HTML attachment was scrubbed...
>     URL:
>     <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20190622/aae9eaba/attachment-0001.html>
>
>     ------------------------------
>
>     Message: 3
>     Date: Sat, 22 Jun 2019 14:54:40 +0100
>     From: Ben Hourahine <benjamin.hourahine at strath.ac.uk
>     <mailto:benjamin.hourahine at strath.ac.uk>>
>     To: dftb-plus-user at mailman.zfn.uni-bremen.de
>     <mailto:dftb-plus-user at mailman.zfn.uni-bremen.de>
>     Subject: Re: [DFTB-Plus-User] Possible bug in DFTB+ with MPI
>     Message-ID: <16da7e6a-92dd-2184-7a98-067e751cdb14 at strath.ac.uk
>     <mailto:16da7e6a-92dd-2184-7a98-067e751cdb14 at strath.ac.uk>>
>     Content-Type: text/plain; charset="utf-8"
>
>     Hello David,
>
>     thoughts so far:
>
>     1) Which blas are you using? Sometimes openBlas steals extra
>     threads, is
>     your top output shown with thread reporting enabled (H)?
>
>     2) Are you definitely getting the OMP_NUM_THREADS shell variable
>     to each
>     of the instances?
>
>     3) If this is related to the i-PI mailing list question, is there any
>     reason not to run 8 serial instances with 1 thread each? The MPI
>     parallelism overhead on just one process is probably not worth it.
>
>     Regards
>
>     Ben
>
>     On 22/06/2019 14:41, David Furman wrote:
>     > Hi all,
>     > I have a question concerning the running of DFTB+ with MPI. There
>     > seems to be a problem when running more than one DFTB+ instance
>     (with
>     > MPI parallelization) on the same system.
>     >
>     > When I run 4 instances (separate folders) of dftb+ with 2 cores each
>     > (i.e. 8 cores in total):
>     > mpirun.openmpi -np 2 dftb+, each %CPU usage drops to 50%.
>     >
>     > This is the 'top' output:
>     > ========================================================
>     >  35426 user    20   0  423384  40404  18248 R  50.5  0.0  
>     0:37.69 dftb+
>     >  35434 user    20   0  423080  39720  18248 R  50.2  0.0  
>     0:05.70 dftb+
>     >  35396 user    20   0  422044  38660  17644 R  49.8  0.0  
>     1:21.35 dftb+
>     >  35402 user    20   0  423384  40460  18304 R  49.8  0.0  
>     1:09.54 dftb+
>     >  35403 user    20   0  422044  38872  17848 R  49.8  0.0  
>     1:16.15 dftb+
>     >  35427 user    20   0  422044  39048  18032 R  49.8  0.0  
>     0:32.93 dftb+
>     >  35435 user    20   0  421732  38164  17716 R  49.8  0.0  
>     0:09.07 dftb+
>     >  35395 user    20   0  423384  40360  18204 R  49.5  0.0  
>     1:13.18 dftb+
>     > =========================================================
>     >
>     > Whereas, when I run one instance with 8 cores, the efficiency is
>     100%
>     > as expected:
>     >
>     > =========================================================
>     >  35837 user    20   0  441868  35196  19112 R 100.3  0.0  
>     0:04.90 dftb+
>     >  35828 user    20   0  442580  35676  19120 R 100.0  0.0  
>     0:04.83 dftb+
>     >  35829 user    20   0  442544  36024  19432 R 100.0  0.0  
>     0:04.91 dftb+
>     >  35830 user    20   0  441912  35356  19360 R 100.0  0.0  
>     0:04.91 dftb+
>     >  35831 user    20   0  441688  35048  19304 R 100.0  0.0  
>     0:04.90 dftb+
>     >  35833 user    20   0  441904  35656  19500 R 100.0  0.0  
>     0:04.91 dftb+
>     >  35841 user    20   0  441516  34824  19076 R 100.0  0.0  
>     0:04.89 dftb+
>     >  35843 user    20   0  441224  34456  18868 R  99.7  0.0  
>     0:04.89 dftb+
>     > =========================================================
>     >
>     > I run both cases with OMP_NUM_THREADS=1.
>     > with open-mpi 1.10.2 and gcc 5.4.0.
>     >
>     > Could anyone give a hint about what is wrong?
>     >
>     >
>     > _______________________________________________
>     > DFTB-Plus-User mailing list
>     > DFTB-Plus-User at mailman.zfn.uni-bremen.de
>     <mailto:DFTB-Plus-User at mailman.zfn.uni-bremen.de>
>     >
>     https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
>
>     -- 
>             Dr.  B.  Hourahine,  Senior  lecturer
>                 SUPA, Department  of  Physics,
>                  University  of  Strathclyde,
>                   John  Anderson  Building,
>               107 Rottenrow, Glasgow G4 0NG, UK.                   
>       +44 141 548 2325, benjamin.hourahine at strath.ac.uk
>     <mailto:benjamin.hourahine at strath.ac.uk>
>
>     The  Department is  a partner  in SUPA,  the Scottish
>                 Universities Physics Alliance
>
>     The University  of Strathclyde  is a  charitable body,
>            registered in Scotland, number SC015263
>
>     -------------- next part --------------
>     An HTML attachment was scrubbed...
>     URL:
>     <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20190622/bf90b2fb/attachment.html>
>
>     ------------------------------
>
>     Subject: Digest Footer
>
>     _______________________________________________
>     DFTB-Plus-User mailing list
>     DFTB-Plus-User at mailman.zfn.uni-bremen.de
>     <mailto:DFTB-Plus-User at mailman.zfn.uni-bremen.de>
>     https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
>
>     ------------------------------
>
>     End of DFTB-Plus-User Digest, Vol 58, Issue 11
>     **********************************************
>
>
>
> -- 
> David Furman, PhD, MRSC**Herchel Smith Research Fellow Department of
> Chemistry University of Cambridge Lensfield Road CB2 1EW
>
> _______________________________________________
> DFTB-Plus-User mailing list
> DFTB-Plus-User at mailman.zfn.uni-bremen.de
> https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user

-- 
        Dr.  B.  Hourahine,  Senior  lecturer
            SUPA, Department  of  Physics,
             University  of  Strathclyde,
              John  Anderson  Building,
          107 Rottenrow, Glasgow G4 0NG, UK.                    
  +44 141 548 2325, benjamin.hourahine at strath.ac.uk

The  Department is  a partner  in SUPA,  the Scottish
            Universities Physics Alliance

The University  of Strathclyde  is a  charitable body,
       registered in Scotland, number SC015263

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20190622/6c950d2c/attachment-0001.html>


More information about the DFTB-Plus-User mailing list