Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly  
We are NATO-free zone

Re^5: Regarding FIREFLY-8.1.0 running speeds on Cluster with IntelMPI

Alex Granovsky

Dear Siddheshwar,

I'd suggest you to use -o command line option to redirect output
to file. Are you using it? If you are not, then most likely output
gets redirected by the batch system on the cluster. This way batch
system may wait for job completion before creating output file for
you. With -o option you get output in a real time.

>Alex Sir, do I need to have MULTIPLE executables of FIREFLY, if I need to run several scripts parallely?

No you do not need to have multiple copies. A single copy is enough.

Kind regards,
Alex Granovsky

On Tue Apr 15 '14 7:02am, Siddheshwar Chopra wrote
>Dear Alex,
>No Sir. It is not that I didn't get the output. I get the o/p of short programs. The Temp Dir has root access and ONLY the o/p that completes moves in to the user folder. The configuration is fine, it is just that the program is taking more time than my expectation.
>Alex Sir, do I need to have MULTIPLE executables of FIREFLY, if I need to run several scripts parallely?


>On Tue Apr 15 '14 0:58am, Alex Granovsky wrote
>>Dear Siddheshwar,

>>if you have not gotten output files thus far then you most likely
>>misconfigured or misused Firefly on the cluster.

>>Kind regards,
>>Alex Granovsky
>>On Sat Apr 12 '14 6:57am, Siddheshwar Chopra wrote
>>>Dear Alex,
>>>I would need some time to provide the o/p from cluster. Actually it is giving me o/p ONLY when it is completed. But I seem to add less walltime everytime..And that my concern.. It needs more than 2 days atleast..So I slowly increasing it.


>>>On Fri Apr 11 '14 4:33pm, Alex Granovsky wrote

>>>>could you please send me two outputs, one when running on cluster
>>>>and second from your laptop.

>>>>Kind regards,
>>>>Alex Granovsky
>>>>On Fri Apr 11 '14 12:35pm, Siddheshwar Chopra wrote
>>>>>Dear Alex and Firefly users,
>>>>>I am working with Intel MPI version of Firefly-8.1.0 on a cluster. I have run say a OPT+HESSIAN job of around 45 atoms of Silicon on my laptop (with i3, 2.2GHz,2GB DDR3 RAM) and also on a cluster (with 2 nodes and ppn=16 (max support),64GB RAM).
>>>>>The program ends in around 7 days on my laptop, but the cluster also takes more than 3 days for the same. The commands in the i/p file are:

>>>>> $SYSTEM MEMORY=34459319 aoints=dist TIMLIM=5295600 KDIAG=0 $END
>>>>> $CONTRL MAXIT=10000 MOLPLT=.t. PLTORB=.t. $END
>>>>> $STATPT opttol=10E-8 NSTEP=200 Method=GDIIS HSSEND=.TRUE. $END
>>>>> $SMP CALL64=.t. HTTFIX=.f. $END
>>>>> $p2p p2p=.t. mxbuf=2048 dlb=.t. $END
>>>>> $SCF dirscf=.t. fdiff= .f. diis=.t. $END  
>>>>> $DFT HFX(1)=0.15 LMAX=41 NRAD=99 $end
>>>>> $ZMAT DLC=.t. AUTO=.t. $END
>>>>> $SCF NCONV=8 $END
>>>>>Could anyone help me out in explaining what wrong is being done or how to optimize the program speeds on cluster? It has become very essential to understand this issue to progress further.


[ Previous ] [ Next ] [ Index ]           Tue Apr 15 '14 11:07am
[ Reply ] [ Edit ] [ Delete ]           This message read 588 times