[gmx-users] Normal mode segmentation fault - memory problem ?

[David van der Spoel]

Den 2019-07-24 kl. 15:10, skrev Marlon Sidore:
> I would like a second opinion on the cause of this segfault.
> 
> I have a protein in a vacuum. All atom. After extensive minimization, I
> produced a matrix .mtx.
> 
> Then, on the cluster, a simple gmx_mpi nmeig -f normal.mtx -s normal.tpr
> leads to:
> Reading double precision matrix generated by GROMACS 2018.6
> Sparse matrix storage format, nrow=27579, ncols=27579
> [n2534:10432:0] Caught signal 11 (Segmentation fault)
> ==== backtrace ====
>   2 0x000000000006ba2c mxm_handle_error()
>   /var/tmp/OFED_topdir/BUILD/mxm-3.7.3111/src/mxm/util/debug/debug.c:641
>   3 0x000000000006bf7c mxm_error_signal_handler()
>   /var/tmp/OFED_topdir/BUILD/mxm-3.7.3111/src/mxm/util/debug/debug.c:616
>   4 0x0000000000036280 killpg()  ??:0
>   5 0x0000000000642091
> _ZN17_INTERNAL3a3192c818nma_sparse_hessianEP16gmx_sparsematrixiPK10t_topologyRKSt6vectorImSaImEEiPfSA_()
>   /scratch/DEFAUT/akira/
> akira/installs/occigen/applications/gromacs/2018.6/intel/18.1/openmpi/intel/2.0.4/gromacs-2018.6/src/gromacs/gmxana/gmx_nmeig.cpp:220
>   6 0x0000000000642091 gmx_nmeig()
>   /scratch/DEFAUT/akira/akira/installs/occigen/applications/gromacs/2018.6/intel/18.1/openmpi/intel/2.0.4/gromacs-201
> 8.6/src/gromacs/gmxana/gmx_nmeig.cpp:471
>   7 0x0000000000431cb9
> _ZN3gmx12_GLOBAL__N_122CMainCommandLineModule3runEiPPc()
>   /scratch/DEFAUT/akira/akira/installs/occigen/applications/gromacs/2018
> .6/intel/18.1/openmpi/intel/2.0.4/gromacs-2018.6/src/gromacs/commandline/cmdlinemodulemanager.cpp:133
>   8 0x0000000000431cb9 _ZN3gmx24CommandLineModuleManager3runEiPPc()
>   /scratch/DEFAUT/akira/akira/installs/occigen/applications/gromacs/2018.6/intel/18.
> 1/openmpi/intel/2.0.4/gromacs-2018.6/src/gromacs/commandline/cmdlinemodulemanager.cpp:589
>   9 0x000000000040f788 main()
>   /scratch/DEFAUT/akira/akira/installs/occigen/applications/gromacs/2018.6/intel/18.1/openmpi/intel/2.0.4/gromacs-2018.6/s
> rc/programs/gmx.cpp:60
> 10 0x00000000000223d5 __libc_start_main()  ??:0
> 11 0x000000000040f5e9 _start()  ??:0
> ===================
> 
> Either I did something wrong while producing the matrix or I'm doing
> something wrong on the nmeig. S
> 
> Should I have done something to restrict the analysis on the backbone
> (somehow) if it's a memory problem ? I'm still allocacating 100G to
> this analysis, I naively thought it was enough.
27500^2 * 8 equals 6 Gb. Should not be a memory problem.
Have you tried using
-end 82737
In that case the full matrix will be used. It may be slow though.
> 
> Best,
> 
> Marlon Sidore
> 
> PhD - Post-doctoral fellow
> Institut d’Électronique et des Systèmes
> 


-- 
David van der Spoel, Ph.D., Professor of Biology
Head of Department, Cell & Molecular Biology, Uppsala University.
Box 596, SE-75124 Uppsala, Sweden. Phone: +46184714205.
http://www.icm.uu.se