[gmx-users] Query regarding Domain decomposition

Mark Abraham mark.j.abraham at gmail.com
Wed Jun 25 14:23:45 CEST 2014


On Jun 25, 2014 8:15 AM, "suhani nagpal" <suhani.nagpal at gmail.com> wrote:
>
> Greetings
>
> I have been trying to run a few set of simulations using high number of
> processors.
>
> Using the tutorial -
>
http://compchemmpi.wikispaces.com/file/view/Domaindecomposition_KKirchner_27Apr2012.pdf
>
> I have done calculations to evaluate the set of nodes which would be
> optimal for the protein.
>
>
> So the all the files are generated, but error occurs and the trajectory
> files remain empty with no error mentioned in the log file.

Hard to say. The problem could be anywhere, since we don't yet know when
mdrun does work...

> Number of nodes to be used in multiple of 16
>
> box in x and y dimension 8 nm
>
>
>
> In the error file,
>
>
> Reading file 400K_SIM2.tpr, VERSION 4.5.5 (single precision)

Why use a slow, old version if you want parallel performance?

> Note: file tpx version 73, software tpx version 83

You should prefer to use grompp whose version matches mdrun.

> The number of OpenMP threads was set by environment variable
> OMP_NUM_THREADS to 1
> Using 320 MPI processes
>
> NOTE: The load imbalance in PME FFT and solve is 116%.
>       For optimal PME load balancing
>       PME grid_x (54) and grid_y (54) should be divisible by #PME_nodes_x
> (140)
>       and PME grid_y (54) and grid_z (54) should be divisible by
> #PME_nodes_y (1)
>
>
>
> mdp file for reference
>
> ; Bond parameters
> continuation    = yes           ; Restarting after NPT
> constraint_algorithm = lincs    ; holonomic constraints
> constraints     = all-bonds     ; all bonds (even heavy atom-H bonds)
> constrained
> lincs_iter      = 1             ; accuracy of LINCS
> lincs_order     = 4             ; also related to accuracy
> ; Neighborsearching
> ns_type         = grid          ; search neighboring grid cells
> nstlist         = 5             ; 10 fs
> rlist           = 1.0           ; short-range neighborlist cutoff (in nm)
> rcoulomb        = 1.0           ; short-range electrostatic cutoff (in nm)
> rvdw            = 1.0           ; short-range van der Waals cutoff (in nm)
> ; Electrostatics
> coulombtype     = PME           ; Particle Mesh Ewald for long-range
> electrostatics
> pme_order       = 4             ; cubic interpolation
> fourierspacing  = 0.16          ; grid spacing for FFT
> ; Temperature coupling is on
> tcoupl          = nose-hoover   ; nose-hoover coupling
> tc-grps         = Protein Non-Protein   ; two coupling groups - more
> accurate
> tau_t           = 0.2   0.2     ; time constant, in ps
> ref_t           = 400   400     ; reference temperature, one for each
> group, in K
> ; Pressure coupling is off
> pcoupl          = no            ;
> ; Periodic boundary conditions
> pbc             = xyz           ; 3-D PBC
> ; Dispersion correction
> DispCorr        = EnerPres      ; account for cut-off vdW scheme
> ; Velocity generation
> gen_vel         = yes           ; assign velocities from Maxwell
> distribution
> gen_temp        = 400           ; temperature for Maxwell distribution
> gen_seed        = -1            ; generate a random seed
>
>
> Kindly help.
>
> I have to run simulations at 250 to 300 processors.

Maybe. You can't efficiently parallelize an algorithm over arbitrary
amounts of hardware. You need 100-1000 atoms per core, depending on
hardware, simulation settings and GROMACS version.

Mark

> --
> 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.


More information about the gromacs.org_gmx-users mailing list