[gmx-users] delay at start

[Michael Brunsteiner]

Hi,

thanks Szilard for the reply, 
below I include what i get to stdout/stderr and the mdp (log file is
too long) obviously the timings are not accurate due to buffering
but to me it looks as if Gromacs spent reproducibly about 2 minutes
running on ONE thread, and only then switches to use 4. This switch
happens at the time when it starts doing the test runs to establish
the optimal pme grid. these test runs only take a couple of seconds
(about 3000 steps = 6ps of a job which on this architecture does
around 220 ns/d) from then on-wards everything is normal and as expected...

In the log file after job-start it gets quickly to the point
where it says:

"Will do PME sum in reciprocal space for electrostatic interactions."

followed by some lit reference. then is stops there for about 2 minutes
and only then carries on saying:

Using a Gaussian width (1/beta) of 0.384195 nm for Ewald

in these 2 minutes the job runs only on one thread, later it switches to 4.
This is on a work station with Intel i7-3820 at 3.60GHz + one Nvidia 1060 GTX 6Gb
my gmx is 2018.4 on a recently updated debian testing/buster


=======================
stderr/stdout: job starts at  01:26:14
=======================
                      :-) GROMACS - gmx mdrun, 2018.4 (-:

                            GROMACS is written by:
     Emile Apol      Rossen Apostolov      Paul Bauer     Herman J.C. Berendsen
    Par Bjelkmar    Aldert van Buuren   Rudi van Drunen     Anton Feenstra  
  Gerrit Groenhof    Aleksei Iupinov   Christoph Junghans   Anca Hamuraru   
 Vincent Hindriksen Dimitrios Karkoulis    Peter Kasson        Jiri Kraus    
  Carsten Kutzner      Per Larsson      Justin A. Lemkul    Viveca Lindahl  
  Magnus Lundborg   Pieter Meulenhoff    Erik Marklund      Teemu Murtola   
    Szilard Pall       Sander Pronk      Roland Schulz     Alexey Shvetsov  
   Michael Shirts     Alfons Sijbers     Peter Tieleman    Teemu Virolainen 
 Christian Wennberg    Maarten Wolf   
                           and the project leaders:
        Mark Abraham, Berk Hess, Erik Lindahl, and David van der Spoel

Copyright (c) 1991-2000, University of Groningen, The Netherlands.
Copyright (c) 2001-2017, The GROMACS development team at
Uppsala University, Stockholm University and
the Royal Institute of Technology, Sweden.
check out http://www.gromacs.org for more information.

GROMACS is free software; you can redistribute it and/or modify it
under the terms of the GNU Lesser General Public License
as published by the Free Software Foundation; either version 2.1
of the License, or (at your option) any later version.

GROMACS:      gmx mdrun, version 2018.4
Executable:   /home/michael/local/gromacs-2018.4-bin-cuda10/bin/gmx
Data prefix:  /home/michael/local/gromacs-2018.4-bin-cuda10
Working dir:  /home/michael/test/test-start-job-delay
Command line:
  gmx mdrun -v -deffnm testd

Reading file testd.tpr, VERSION 2018.4 (single precision)

stops here until about 01:26:32 ========================================

Changing nstlist from 50 to 100, rlist from 1.2 to 1.236

Using 1 MPI thread
Using 4 OpenMP threads 

1 GPU auto-selected for this run.
Mapping of GPU IDs to the 2 GPU tasks in the 1 rank on this node:
  PP:0,PME:0

stops here until 01:28:46 ==============================================

starting mdrun 'system'
1000000 steps,   2000.0 ps.
^Mstep  200: timed with pme grid 44 48 48, coulomb cutoff 1.200: 332.1 M-cycles
^Mstep  400: timed with pme grid 42 42 44, coulomb cutoff 1.303: 362.5 M-cycles

stops here until 01:28:47 ===========================================

^Mstep  600: timed with pme grid 36 40 40, coulomb cutoff 1.456: 390.1 M-cycles
^Mstep  800: timed with pme grid 40 40 40, coulomb cutoff 1.394: 365.9 M-cycles
^Mstep 1000: timed with pme grid 40 40 42, coulomb cutoff 1.368: 355.7 M-cycles
^Mstep 1200: timed with pme grid 40 42 42, coulomb cutoff 1.328: 343.1 M-cycles
^Mstep 1400: timed with pme grid 40 42 44, coulomb cutoff 1.310: 340.4 M-cycles
^Mstep 1600: timed with pme grid 42 42 44, coulomb cutoff 1.303: 337.4 M-cycles
^Mstep 1800: timed with pme grid 42 44 44, coulomb cutoff 1.268: 325.7 M-cycles
^Mstep 2000: timed with pme grid 42 44 48, coulomb cutoff 1.248: 316.7 M-cycles
^Mstep 2200: timed with pme grid 44 48 48, coulomb cutoff 1.200: 295.9 M-cycles
^Mstep 2400: timed with pme grid 42 44 44, coulomb cutoff 1.268: 324.1 M-cycles
^Mstep 2600: timed with pme grid 42 44 48, coulomb cutoff 1.248: 317.3 M-cycles
^Mstep 2800: timed with pme grid 44 48 48, coulomb cutoff 1.200: 294.8 M-cycles
^M              optimal pme grid 44 48 48, coulomb cutoff 1.200

stops here until 01:28:50 ==========================================



mdp file:

integrator               = md
dt                       = 0.002
nsteps                   = 1000000
comm-grps                = System
nstcomm                  = 50
;
nstxout                  = 25000
nstvout                  = 0
nstfout                  = 0
nstlog                   = 25000
nstenergy                = 25000
;
nstlist                  = 50
ns_type                  = grid
pbc                      = xyz
rlist                    = 1.2
cutoff-scheme            = Verlet
;
coulombtype              = PME
rcoulomb                 = 1.2
vdw_type                 = cut-off 
rvdw                     = 1.2
;
constraints              = h-bonds
;
tcoupl            = v-rescale
tau-t             = 0.1
ref-t             = 313.0
tc-grps           = System
;
pcoupl           = berendsen
pcoupltype       = anisotropic
tau-p            = 2.0
compressibility  = 4.5e-5 4.5e-5 4.5e-5 0 0 0
ref-p            = 1 1 1 0 0 0