[Wien] Will mpi parallelization need .machines file ?

[venky ch]

Dear Wien2k users,

I would like to understand the difference between the k-point
parallelization and mpi parallelization. Simply, I can mention two
different .machines file for respective parallelization as given below,

For example, if I have 100 k-points and 50 CPUs , the .machines files would
be as follows

======================= k-point parallization; 100 k-points ========

The k-points are finely distributed over one CPU. So, number of lines are
equal to #of CPUs

50 CPUs =  50 lines

2:machine1:1            (machine1: 1 processor)
2:machine2:1
2:machine2:1
2:machine2:1
.
.
.
.
2:machine2:1 (machine2: 13 processors)
2:machine3:1
2:machine3:1 (machine3: 2 processors)
2:machine4:1
.
.
.
.
2:machine4:1            (machine4: 20 processors)
2:machine5:1
2:machine5:1
.
.
.
.
2:machine5:1
2:machine5:1   (machine5: 15 prcessors)

======================================

======================= mpi parallization========

The k-points are coarsely distributed over one node. So, for a given node
an entire number of CPUs have been used. The total k-points (100) are
distributed among the entire nodes

48:machine1:24
48:machine2:24
4:machine3:2

 ===========================

As shown in above, one can switch between k-point parallelization and mpi
parallelization depending on the arrangement in the .machines file. Anyone,
kindly correct my view on this.

Further, I would like to ask, what if one can't get the list of
assigned nodes to the parallel job. In that case, how we can generate the
.machines file for parallel calculations. I mean, Is it possible to do the
parallel calculation without a .machines file or Is there any universal way
of writing a .machines file that can fit into any cluster (even if one
doesn't know the details of the nodes assigned to the parallel job).


These clarifications are very much helpful to me.

thanks

Venkatesh
Postdoctoral Fellow,
Department of Instrumentation and Applied Physics
IISc Bangalore, India
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20210806/0f797965/attachment.htm>