Introduction to OpenMP and Open MPI16 Aug 2015
Computers today are getting faster and more powerful everyday. In the past, the answer to faster computers was to increase the clock speed (e.g. 2.0 GHz to 2.5 GHz) of the computer. This, however, also increases the power consumption of the computer. This increasing power consumption is not sustainable because with more power, the most heat that is generated and even more power is needed to cool down the processors. Thus, faster hardware is not the answer.
The solution for faster computers has now shifted from improving the hardware to optimizing the software. Instead of just one very fast processor, there would be smaller but multiple processors doing the work. Thus, parallel processing is needed and frameworks such as OpenMP and Open MPI were created to facilitate parallel processing. Additionally, recently the White House blogged about establishing the National Strategic Computing Initiative (NSCI) to lead the U.S. in advanced computing.
Just briefly, OpenMP (which stands for Open Multi-Processing) is an API which supports multi-platform shared memory multiprocessing programming in C, C++, and Fortran (1). What this means is that you can parallize your programs on a shared memory machine (e.g. your desktop or laptop with multiple processing cores).
Open MPI, similarly, enables the user to parallize their programs but in a different fashion. Open MPI is an implementation of MPI, which stands for “message-passing interface.” MPI is “a message-passing library interface specification.” More information can be found in the “MPI: A Message-Passing Interface Standard” documentation (2).
The reason Open MPI is different from OpenMP is because Opne MPI works on distributed memory systems, i.e. supercomputer or cluster computers. Thus, message passing makes more sense when you think about it in terms of a cluster of computers.
MPI is a way of doing computing, not necessarily the actually implementation. Open MPI is just one implementation of MPI. Here are some other implementations of MPI in R and Python:
To reiterate, MPI is a way to parallize your functions and programs, but a language. So any implementation must create a library in order to interface with the hardware you are trying to optimize.