From: Marcelo C. R. Melo (melomcr_at_gmail.com)
Date: Wed Jan 02 2019 - 09:33:25 CST
We have repeatedly ran NAMD QM/MM on local clusters here in the US and in
Germany, and even on Blue Waters (a somewhat large supercomputer), both
using ORCA and MOPAC (and some people tried with other codes for certain
tests and benchmarks).
I see Jim is helping you with very advanced and specific configuration
options (the type of thing that most people using supercomputers don't even
know about), so you should try to get that working by following his
suggestions and checking the specific documentation of the cluster you are
Other than that, you should also remember to benchmark you simulations to
know where the bottleneck is.
For example, in our case, running our simulations of ~250 QM atoms with
MOPAC was more than 3 times faster in out local computers, but running in
Blue Waters we could use hundreds of computers at once, making the overall
process of running multiple replicas in different conditions faster. On
supercomputers, you should win in the *distribution* of QM/MM calculations,
not necessarily on the speed of individual calculations, especially when
using codes that don't parallelize on multiple nodes, like MOPAC.
1) Try running using the qmBaseDir on RAM (/dev/shm) and on a local disk.
If you don't see a difference, you will know the CPU is limiting you (which
is to be expected in larger systems).
2) If your system is really small, running MOPAC in multiple cores may not
be helping you much depending on the hardware. Try using threads=1, 2, 4,
8, 16. You should see where you stop gaining speed.
In my experience, if you feel you are not getting an "efficient" use of the
cluster, finding out where is the limitation is the best way to figure out
how to best use the cluster resources.\
--- Marcelo Cardoso dos Reis Melo PhD Candidate Luthey-Schulten Group University of Illinois at Urbana-Champaign crdsdsr2_at_illinois.edu +1 (217) 244-5983 On Mon, 31 Dec 2018 at 01:34, Francesco Pietra <chiendarret_at_gmail.com> wrote: > Hi all: > As I was unable to run efficiently QM-MM with NAMD-MOPAC on a cluster > (recent exchange of mail on this foru: I was unable to run faster than on a > laptop those simulations), may I ask about success, on whichever cluster, > where I failed? > Thanks > francesco pietra >
This archive was generated by hypermail 2.1.6 : Tue Dec 31 2019 - 23:20:25 CST