High Performance Computing - Charles Severance [112]
The languages are designed to be portable, but if the vendor of your particular scalable computer doesn't support the language variant in which you have chosen to write your application, then it isn't portable. Even if the vendor has your language available, it may not be tuned to generate the best code for their architecture.
One solution is to purchase your compilers from a third-party company such as Pacific Sierra or Kuck and Associates. These vendors sell one compiler that runs across a wide range of systems. For users who can afford these options, these compilers afford a higher level of portability.
One of the fundamental issues is the chicken-and-egg problem. If users don't use a language, vendors won't improve the language. If all the influential users (with all the money) use message passing, then the existence of an excellent HPF compiler is of no real value to those users.
The good news is that both FORTRAN 90 and HPF provide one road map to portable scalable computing that doesn't require explicit message passing. The only question is which road we users will choose.
4.2. Message-Passing Environments
Introduction*
A message-passing interface is a set of function and subroutine calls for C or FORTRAN that give you a way to split an application for parallel execution. Data is divided and passed out to other processors as messages. The receiving processors unpack them, do some work, and send the results back or pass them along to other processors in the parallel computer.
In some ways, message passing is the “assembly language” of parallel processing. You get ultimate responsibility, and if you are talented (and your problem cooperates), you get ultimate performance. If you have a nice scalable problem and are not satisfied with the resulting performance, you pretty much have yourself to blame. The compiler is completely unaware of the parallel aspects of the program.
The two most popular message-passing environments are parallel virtual machine (PVM) and message-passing interface (MPI). Most of the important features are available in either environment. Once you have mastered message passing, moving from PVM to MPI won’t cause you much trouble. You may also operate on a system that provides only a vendor-specific message-passing interface. However, once you understand message passing concepts and have properly decomposed your application, usually it’s not that much more effort to move from one message-passing library to another.[72]
Parallel Virtual Machine*
The idea behind PVM is to assemble a diverse set of network-connected resources into a “virtual machine.” A user could marshal the resources of 35 idle workstations on the Internet and have their own personal scalable processing system. The work on PVM started in the early 1990s at Oak Ridge National Labs. PVM was pretty much an instant success among computer scientists. It provided a rough framework in which to experiment with using a network of workstations as a parallel processor.
In PVM Version 3, your virtual machine can consist of single processors, shared-memory multiprocessors, and scalable multiprocessors. PVM attempts to knit all of these resources into a single, consistent, execution environment.
To run PVM, you simply need a login account on a set of network computers that have the PVM software installed. You can even install it in your home directory. To create your own personal virtual machine, you would create a list of these computers in a file:
% cat hostfile
frodo.egr.msu.edu
gollum.egr.msu.edu
mordor.egr.msu.edu
%
After some nontrivial machinations with paths and environment variables, you can start the PVM console:
% pvm hostfile
pvmd already running.
pvm> conf
1 host, 1 data format