setting number of processes for MPI.COMM_WORLD

[Alemu Tadesse]

Hello, 

I was developing a parallel application using mpi4py. the application uses large amount of GPS data and i am trying to have two parallelization phase. the first step will be embarrassingly parallel for each group of  GPS data and then on the second phase i will do K-Means clustering . i want to parallelize the K-Means clustering with mpi4py but i couldn't set parallelism degree by myself , i want to set different parallelism degree for each phase.

could anyone give me some suggestions ? 

Kindly,

Alemu Tadesse

[Lisandro Dalcin]

In MPI, you cannot change the number of processes in MPI.COMM_WORLD.

Maybe you could try to use dynamic process management, i.e.
`comm.Spawn()`, there are some examples under the `demo/` directory.

For the embarrassingly parallel part, you may use the new
mpi4py.futures, which is implemented with `comm.Spawn()` (you need to
install the in-development mpi4py from git).


--
Lisandro Dalcin
============
Research Scientist
Computer, Electrical and Mathematical Sciences & Engineering (CEMSE)
Extreme Computing Research Center (ECRC)
King Abdullah University of Science and Technology (KAUST)
http://ecrc.kaust.edu.sa/

4700 King Abdullah University of Science and Technology
al-Khawarizmi Bldg (Bldg 1), Office # 0109
Thuwal 23955-6900, Kingdom of Saudi Arabia
http://www.kaust.edu.sa

Office Phone: +966 12 808-0459