[gmx-users] comparison of reaction field and PME

[Bert de Groot]

--Yuguang Mu wrote:
> 
> Dear Users,
> 
> First of all thank you all for the answers for the parallel problem.
> Now we have scceed in running parallel jobs using LAM version of MPI.
> 
> Here I found another thing:
> 
> I check two nearlly identical simulations of peptides in DMSO
> each is 20 ps long.
> 
> The other difference is the treatment of electro-static interactions
> one  using reaction field (cut -off  )
> rlist               =  1.2
> rcoulomb            =  1.2
> rvdw                =  1.2
> coulombtype         =  Reaction-Field
> 
> Aother one is PME.
> rlist               =  1.2
> rcoulomb            =  1.2
> rvdw                =  1.2
> coulombtype         =  PME
> fourierspacing      =  0.1
> pme_order           =  4
> optimize_fft        =  yes
> 
> The simulation results are all reasonable, no problem.
> 
> But the CPU time used by PME are nearlly 4 times longer that CUT-OFF.
> 
> here is the output for CUT-OFF :
>                NODE (s)   Real (s)      (%)
>        Time:    331.000    331.000    100.0
>                        5:31
>                (Mnbf/s)   (MFlops) (ps/NODE hour) (NODE hour/ns)
> Performance:     10.283    477.522    217.523      4.597
> 
> and here is PME
>                NODE (s)   Real (s)      (%)
>        Time:   1294.000   1294.000    100.0
>                        21:34
>                (Mnbf/s)   (MFlops) (ps/NODE hour) (NODE hour/ns)
> Performance:      2.631    451.218     55.641     17.972
> 
> Do you think it is normal?
> Or further optimize the FFTW codes is necassary?
> 

as written in the manual, optimising the fft only makes sense for long simulations
(the fft optimisation is done only once, at the beginning, and can take quite some time).
I bet most of the CPU time in the PME simulation (20ps!) was spent on optimising the fft.
BTW I haven't seen any aplycation that gained much from the fft optimisation.


Bert

____________________________________________________________________________
Dr. Bert de Groot

Max Planck Institute for Biophysical Chemistry
Theoretical molecular biophysics group
Am Fassberg 11 
37077 Goettingen, Germany

tel: +49-551-2011306, fax: +49-551-2011089

email: bgroot at gwdg.de
http://www.mpibpc.gwdg.de/abteilungen/071/bgroot
____________________________________________________________________________