Hi Elvis and others,
I also did not find a way to separate the two calculations. What i am
saying is:
1. Is such a separation possible by making a some changes in source
code and recompiling. If it is not too technical i am willing to try.
2. Will it be possible to incorporate such a change in future releases
of AMBER.
I just found an example to illustrate my point. Let me share it.
Following is the result of GB+Entropy calculations on a 350ns run on a ~280
residue protein with 5 threads.
Timing:
Total setup time: 0.000 days
Creating trajectories with cpptraj: 0.026 days
Total calculation time: 2.368 days
Total GB calculation time: 1.587 days
Total quasi-harmonic calculation time: 0.746 days
Statistics calculation & output writing: 0.000 days
Total time taken: 2.394 days
Enthalpy calculation took 1.587 days i.e. double of entropy (.7 days). If
the two can be separated (entropy and enthalpy) then the entropy can be
calculated in .7 days by doubling the number of threads (i know scaling is
not linear so time will be bit more). But it will save .7 day. In fact the
nodes i am using i can give upto 16 threads (its a 2 socket each 8 cores).
But for now it is not possible as it will stall entropy due to RAM
constraints.
I hope some MM(GB/PB)SA/ developer in the mailing list might be able to
help us in this.
Regards
Abhilash
On Sat, Feb 4, 2017 at 1:55 PM, Elvis Martis <elvis.martis.bcp.edu.in>
wrote:
> Hi,
> I don't think you can separate this way for memory intensive jobs and
> those which are not memory intensive in MMPBSA for single run.
> However, I usually do my enthalpy calculations on all snapshots and
> entropy calculations on every 5th snapshot.
> And, also perform enthalpy calculations separately and entropy
> calculations separately (MAX with 8 CPUs for 24 GB RAM).
> Best Regards
>
>
>
> Elvis Martis
> Ph.D. Student (Computational Chemistry)
> at Bombay College of Pharmacy
>
>
> A Kalina, Santacruz [E], Mumbai 400098, INDIA
> W www.elvismartis.in
> Skype. adrian_elvis12
>
>
>
>
> -----Original Message-----
> From: Abhilash J [mailto:md.scfbio.gmail.com]
> Sent: Saturday, February 04, 2017 11:57 AM
> To: AMBER Mailing List <amber.ambermd.org>
> Subject: [AMBER] Number of threads in MMGBSA/MMPBSA calculations.
>
> Hi everyone,
>
> I have observed that enthalpy calculation is not RAM intensive and
> entropy is RAM intensive. When we run a job with both entropy and enthalpy
> calculations, number of threads is dictated by the entropy calculations (as
> RAM of a system is constant. 64 GB in my case) but the time is decided
> primarily by entropy calculations as they tend to slow down due to less
> threads. If you increase the threads enthalpy calculation will be fast but
> entropy calculations will fail due to insufficient RAM. Hence there is
> under utilization of the system. The problem is acute where the
> trajectories are large 250ns or more.
> My question is, is it some how possible to separate the number of
> threads in entropy and enthalpy calculations.
> I find that for long simulations this can lead to difference of a day or
> more.
> This will save me and others like me days of calculation in some cases.
>
> Regards
>
> Abhilash
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Feb 04 2017 - 06:00:03 PST
