[gpaw-users] Increase in process memory for single point calculations

[Gaël Donval]

Hi Chang,

> Hi GPAW users,
> 
> 
> I have encountered a strange problem of increasing process memory
> when I did a few single point calculations in loop. Here is my input
> file:
> 
> 
> from gpaw import *
> from ase.io import *
> from gpaw.utilities import h2gpts
> import numpy as np
> 
> 
> names = ['2co','6co']
> states = ['is','ts','fs']
> tsind = {'2co':6,'6co':4}
> 
> mixer = Mixer(0.1, 5, weight = 100.0)
> 
> def calc():
>     return GPAW(poissonsolver = {'dipolelayer':'xy','eps':1e-12},
>                 kpts = {'density':2},
>                 xc = 'RPBE',
>                 symmetry = {'point_group':False},
>                 mixer = mixer,
>                 eigensolver=Davidson(3),
>                 spinpol = False,
>                 occupations = FermiDirac(0.1),
>                 maxiter = 1000,
> 
>     )
> for name in names:
>     neb_states = read('../%s/neb_all.traj@:'%name)
>     for state in states:
>         if state == 'is': slab = neb_states[0]
>         if state == 'fs': slab = neb_states[-1]
>         else: slab = neb_states[tsind[name]]
>         slab.center(vacuum=5.,axis=2)
>         slab.cell[2][2]+=10.
>         slab.pbc = (1,1,0)
>         slab.set_calculator(calc())
>         slab.calc.set(gpts = h2gpts(0.2, slab.get_cell(), idiv=8),txt
> = '%s_%s_gpaw.out' % (name,state))
>         e_f = slab.get_potential_energy()
>         f = open('Energy_%s_%s_gpaw.out' % (name,state), 'w')
>         f.write('%.3f\n' % e_f)
>         f.close()
> 
> For GPAW-1.3.0, the output files for these calculations shows:
> 
> 
> b-an01 [~/pfs/co2rr/2018/180812/1.3.0]$ grep -r --include='*out' -i
> 'process memory' -A 1 $(ls -tr *)
> 2co_is_gpaw.out:  Process memory now: 80.45 MiB
> 2co_is_gpaw.out-  Calculator: 148.19 MiB
> --
> 2co_ts_gpaw.out:  Process memory now: 255.31 MiB
> 2co_ts_gpaw.out-  Calculator: 148.19 MiB
> --
> 2co_fs_gpaw.out:  Process memory now: 256.38 MiB
> 2co_fs_gpaw.out-  Calculator: 156.32 MiB
> --
> 6co_is_gpaw.out:  Process memory now: 414.14 MiB
> 6co_is_gpaw.out-  Calculator: 169.99 MiB
> --
> 6co_ts_gpaw.out:  Process memory now: 414.14 MiB
> 6co_ts_gpaw.out-  Calculator: 169.99 MiB
> --
> 6co_fs_gpaw.out:  Process memory now: 414.14 MiB
> 6co_fs_gpaw.out-  Calculator: 170.05 MiB
> 
> For GPAW-1.4.0, it shows:
> 
> 
> b-an01 [~/pfs/co2rr/2018/180812/1.4.0]$ grep -r --include='*out' -i
> 'process memory' -A 1 $(ls -tr *)
> 2co_is_gpaw.out:  Process memory now: 103.06 MiB
> 2co_is_gpaw.out-  Calculator: 108.73 MiB
> --
> 2co_ts_gpaw.out:  Process memory now: 353.70 MiB
> 2co_ts_gpaw.out-  Calculator: 108.73 MiB
> --
> 2co_fs_gpaw.out:  Process memory now: 394.64 MiB
> 2co_fs_gpaw.out-  Calculator: 114.23 MiB
> --
> 6co_is_gpaw.out:  Process memory now: 589.16 MiB
> 6co_is_gpaw.out-  Calculator: 124.54 MiB
> --
> 6co_ts_gpaw.out:  Process memory now: 589.16 MiB
> 6co_ts_gpaw.out-  Calculator: 124.54 MiB
> --
> 6co_fs_gpaw.out:  Process memory now: 606.05 MiB
> 6co_fs_gpaw.out-  Calculator: 124.60 MiB
> 
> It becomes a severe problem when I tried to do more single point
> calculations with different potentials with the Solvated Jellium
> Model calculator since in that case, the growth of process memory
> goes too fast so that the program crashes. Could you help me? Thank 
> you very much!

This looks awfully similar to a problem I had a few weeks ago (though
it could be something else entirely). Could you tell us what MPI
implementation you are using?

In my case, I could narrow down the problem to OpenMPI 3 but I didn't
have time to look further into this. My quick fix at that time was to
revert back to OpenMPI 2. You could try that if that applies to your
case.

Best regards,
Gaël



> 
> 
> Best wishes,
> Chang Liu