MPI technique/best practice documents

[Mitsuru]

Dear experts

Is there nice technical/best-practice books or documents for MPI for efficient parallelization.

Nowadays, such a documents of shared memory parallelization (openMP/TBB,concurrency's etc) are abundant, but lack in distributed-memory arch.

If you know, I hope you tell me.

Kind regards

[lastchance]

Well, you could try
http://wgropp.cs.illinois.edu/usingmpiweb/

I'm afraid that I learnt MPI in bits and pieces on an "as you need it" sort of basis, so I wouldn't be a great guide for a book. There's also lots of people at work that I can ask.

It helps if you know Fortran as well as C/C++.


I didn't know that Boost had MPI (your other thread). I typically use Microsoft MPI on Windows and mpich on linux.
https://docs.microsoft.com/en-us/message-passing-interface/microsoft-mpi
http://www.mpich.org/

[Mitsuru]

Dear lastchance

Thank you for your kind reply, and I would like to refer the books you suggested since it contains whole topics (including numerical algorithm for physics).

> I typically use Microsoft MPI on Windows and mpich on linux.

I currently use Intel MPI as the base library, and utilize boost::mpi in order easily to handle C++ class types. boost::mpi abstracts C++ types to communicate it as it is, by using boost::serialization (C++ class types are converted to byte array).
Particularly, STL container (c.f. std::vector) can be easily handled.

Probably, the problem of my other thread is due to some library dependency. If anyone does not know, I am willing to struggle to resolve it by myself.

I appreciate your helpful comment.

Kind regards