Firefly and PC GAMESS-related discussion club


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



Re^3: Parallel geometry optimization with XMC-QDPT2

Alex Granovsky
gran@classic.chem.msu.su


Dear Dawid,

it seems you are using the old-style, outdated command line to run
Firefly. It is deprecated and was not designed to work in xp mode.

Usually, one needs to use the following command line switches with Firefly:

Firefly -r -p -stdext -i <  fully qualified file name of the input file > -o < fully qualified file name of the output file > -ex < fully qualified path to directory where *.ex files are stored >  -t < fully qualified path to directory which will be used to create temporary folders and to store temporary files >

You just need to add -xp option to these options.

For description of command line options, see page 32 of the current Firefly's manual.

Hope this helps.

Al the best,
Alex Granovsky






On Wed Dec 28 '16 12:05pm, Dawid wrote
--------------------------------------
>Dear Igor,

>Thank you for pointing that out. However, it does not work as it is supposed to. Again for 24 and 48 processors the performance is the same. I use -xp=$NCPUS option, where again $NCPUS is number of processors declared for my batching system. If I use a number for -xp different from number of processors, e.g. -xp=4 for 24 processors I get error like this

> PROCESS     4 FAILED TO OPEN INPUT FILE, ERROR CODE:       21
> PROCESS     8 FAILED TO OPEN INPUT FILE, ERROR CODE:       21
> PROCESS    12 FAILED TO OPEN INPUT FILE, ERROR CODE:       21
> PROCESS    16 FAILED TO OPEN INPUT FILE, ERROR CODE:       21
> PROCESS    20 FAILED TO OPEN INPUT FILE, ERROR CODE:       21

>Similarly for other combinations there are some processes that fail to open input file even though it is definitely present in both working and scratch directories. Where does it come from?

>Best regards,
>Dawid

>On Sat Dec 24 '16 8:47pm, Igor Polyakov wrote
>---------------------------------------------
>>Dear Dawid,

>>Please look for the extreme parallel mode (-xp command line option) in the Firefly manual.

>>Best regards, Igor

>>On Fri Dec 23 '16 4:34pm, Dawid wrote
>>-------------------------------------
>>>Dear All,

>>>I use Firefly 8.2.0 Linux/OpenMPI v. 1.8.x, dynamically linked version.
>>>I am currently performing the excited-state geometry optimization
>>>in the first excited state using XMC-QDPT2 method and the 2-roots-SA-CASSCF reference wave function.
>>>I have noticed that whether these calculations run on 8 or 48 processors, the performance is the same, i.e. of walltime
>>>required to do one displacement step in geometry optimization.

>>>Is it what I should expect? Is the xmcqdpt2 module properly
>>>parallelized? I call Firefly with following command

>>>mpirun -np $NCPUS /home/addiw17/firefly820_linux_openmpi_1.8/firefly820 -ex /home/addiw17/firefly820_linux_openmpi_1.8/ > firefly.out 2>&1

>>>where $NCPUS is number of CPUS that I declared for my batching system.

>>>Best regards,
>>>Dawid Grabarek

[ This message was edited on Wed Dec 28 '16 at 5:02pm by the author ]


[ Previous ] [ Next ] [ Index ]           Wed Dec 28 '16 5:02pm
[ Reply ] [ Edit ] [ Delete ]           This message read 181 times