[gmx-users] Estimating the required computer resource for a MD project

Alex alexanderwien2k at gmail.com
Wed Aug 15 18:34:43 CEST 2018


Dear all,
In accessing to a supercomputer facility, I have been asked to estimate how
much computer resource (time and processors) I would need for a MD
simulation project, so, I wonder how I can roughly overestimate what the
project need?
Just to give a hint/example below is the time accounting table given in the
end of the Gromacs *.log file for a 3 ns NpT equilibration for a system of
interest containing 51700 atoms haring 64 ranks (60 for PP + 4  for PME )
on two nodes.
The performance varies between 10 and 20 ns per day for such system.

 Computing:          Num   Num      Call    Wall time         Giga-Cycles
                     Ranks Threads  Count      (s)         total sum    %
-----------------------------------------------------------------------------
 Domain decomp.        60    1      30000     134.862      18610.850   0.5
 DD comm. load         60    1      30000       0.376         51.907   0.0
 DD comm. bounds       60    1      29901       3.861        532.785   0.0
 Send X to PME         60    1    3000001     163.956      22625.842   0.6
 Neighbor search       60    1      30001     211.374      29169.406   0.7
 Comm. coord.          60    1    2970000     378.854      52281.547   1.3
 Force                 60    1    3000001   20351.055    2808429.921  70.1
 Wait + Comm. F        60    1    3000001     961.324     132661.926   3.3
 PME mesh *             4    1    3000001   26174.800     240806.816   6.0
 PME wait for PP *                           1041.450       9581.289   0.2
 Wait + Recv. PME F    60    1    3000001    4491.696     619850.624  15.5
 NB X/F buffer ops.    60    1    8940001     179.639      24790.085   0.6
 Write traj.           60    1        330       0.483         66.585   0.0
 Update                60    1    3000001      44.996       6209.424   0.2
 Constraints           60    1    3000001     204.982      28287.321   0.7
 Comm. energies        60    1     300001      32.308       4458.424   0.1
 Rest                                          58.136       8022.782   0.2
-----------------------------------------------------------------------------
 Total                                      27217.901    4006452.726 100.0
-----------------------------------------------------------------------------
(*) Note that with separate PME ranks, the walltime column actually sums to
    twice the total reported, but the cycle count total and % are correct.
-----------------------------------------------------------------------------
 Breakdown of PME mesh computation
-----------------------------------------------------------------------------
 PME redist. X/F        4    1    6000002    1430.989      13165.027   0.3
 PME spread             4    1    3000001    6457.925      59412.576   1.5
 PME gather             4    1    3000001    4496.692      41369.334   1.0
 PME 3D-FFT             4    1    6000002    9933.436      91387.102   2.3
 PME 3D-FFT Comm.       4    1    6000002    1543.890      14203.705   0.4
 PME solve Elec         4    1    3000001    2298.771      21148.577   0.5
-----------------------------------------------------------------------------

               Core t (s)   Wall t (s)        (%)
       Time:  1741945.673    27217.901     6400.0
                         7h33:37
                 (ns/day)    (hour/ns)
Performance:        9.523        2.520

Thank you.
Regards,
Alex


More information about the gromacs.org_gmx-users mailing list