[gmx-users] Affinity setting for 1/16 threads failed. Version 5.0.2

[Mark Abraham]

Hi,

These are good issues to take up with the support staff of gordon. mdrun
tries to be a good citizen and by default stays out of the way if some
other part of the software stack is already managing process affinity. As
you can see, doing it right is crucial for good performance. But mdrun -pin
on always works everywhere we know about.

Mark

On Wed, Nov 18, 2015 at 3:28 PM Siva Dasetty <sdasett at g.clemson.edu> wrote:

> Dear all,
>
> I am running simulations using version 5.0.2 (default in gordon) and I am
> having a performance loss from 180 ns/day to 8 ns/day compared to the same
> simulations that I previously ran in a different cluster.
>
> In both the clusters I am using a single node and 16 cpus (no gpus) and the
> following is the command line I used:
>
> mdrun_mpi -s <tpr file> -v -deffnm <output file> -nb cpu -cpi <cpt file>
> -append -pin on
>
>
> Following is reported in the log file:
>
>
> WARNING: Affinity setting for 1/16 threads failed.
>
>          This can cause performance degradation! If you think your setting
> are correct, contact the GROMACS developers.
>
>
> I even tried running a simulation without the flag -pin on and there is no
> change in the performance.
>
>
> Are there any other options that I can try to recover the performance?
>
>
>
> Additional Information:
>
>
> The other difference I could see is in the compilers:
>
>
> In gordon (8ns/day):
>
>
> C compiler: /opt/mvapich2/intel/ib/bin/mpicc Intel 13.0.0.20121010
>
> C compiler flags:    -mavx    -std=gnu99 -w3 -wd111 -wd177 -wd181 -wd193
> -wd271 -wd304 -wd383 -wd424 -wd444 -wd522 -wd593 -wd869 -wd981 -wd1418
> -wd1419 -wd1572 -wd1599 -wd2259 -wd2415 -wd2547 -wd2557 -wd3280 -wd3346
> -wd11074 -wd11076  -O3 -DNDEBUG -ip -funroll-all-loops -alias-const
> -ansi-alias
>
> C++ compiler:       /opt/mvapich2/intel/ib/bin/mpicxx Intel 13.0.0.20121010
>
> C++ compiler flags:  -mavx    -w3 -wd111 -wd177 -wd181 -wd193 -wd271 -wd304
> -wd383 -wd424 -wd444 -wd522 -wd593 -wd869 -wd981 -wd1418 -wd1419 -wd1572
> -wd1599 -wd2259 -wd2415 -wd2547 -wd2557 -wd3280 -wd3346 -wd11074 -wd11076
> -wd1782 -wd2282  -O3 -DNDEBUG -ip -funroll-all-loops -alias-const
> -ansi-alias
>
>
> In our cluster (180 ns/day):
>
>
> C compiler: /software/openmpi/bin/mpicc GNU 4.8.1
>
> C compiler flags:    -msse4.1    -Wno-maybe-uninitialized -Wextra -Wno-miss
>
> ing-field-initializers -Wno-sign-compare -Wpointer-arith -Wall -Wno-unused
>
> -Wunused-value -Wunused-parameter  -O3 -DNDEBUG -fomit-frame-pointer -funro
>
> ll-all-loops -fexcess-precision=fast  -Wno-array-bounds
>
> C++ compiler: /software/openmpi/bin/mpicxx GNU 4.8.1
>
> C++ compiler flags:  -msse4.1    -Wextra -Wno-missing-field-initializers -W
>
> pointer-arith -Wall -Wno-unused-function  -O3 -DNDEBUG -fomit-frame-pointer
>
>  -funroll-all-loops -fexcess-precision=fast  -Wno-array-bounds
>
>
>
>
> Thanks in advance for your help,
>
> --
> Siva Dasetty
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
>