QuantumATK Forum

QuantumATK => Scripts, Tutorials and Applications => Topic started by: Dipankar Saha on October 1, 2015, 23:51

Title: Parallel scheduler in Python
Post by: Dipankar Saha on October 1, 2015, 23:51

Hi,
As per the new features in VNL-ATK 2015_
I find that there is_  ' "Parallel scheduler in Python"  which gives a possibility to use MPI directly ............'
__
I wanted to ask...,  for Version 2015.rc1 ....while running a  *.py file through 'mpiexec' command_
will it be the very same as it used to be (previously)?? How, can it be assured that the max. resources are getting utilized...?

Regards_
Dipankar

Title: Re: Parallel scheduler in Python
Post by: Anders Blom on October 2, 2015, 10:04

If you run with mpiexec in 2015 and 2014, the behavior is more or less exactly the same. To get optimal performance you always need to know what you are doing regarding the parallelization, and we are working on better guides, esp. for the new options in 2015.

The ParallelScheduler is an advanced option which requires quite deep knowledge, and is not related to the usual operation of running a script in parallel. It can be used if you for instance inside your script have 4 tasks to be done, then you can parallelize them.

Title: Re: Parallel scheduler in Python
Post by: Dipankar Saha on October 2, 2015, 12:56

Quote from: Anders Blom on October  2, 2015, 10:04

.............. and we are working on better guides, esp. for the new options in 2015.

It would be of great help.....!!  :)

Thanks_
Dipankar