During the last years we have been witnesses to an impressive growth in the calculation power of computational systems. Furthermore, software tools and programming languages have followed these advances providing new techniques and programming paradigms to use the installed hardware in more efficient ways, including the use of parallel programming. Unfortunately, to take full advantage of the power of these computational systems it is necessary, in most cases, to rewrite and optimize our older routines and, more often than not, to attack the problem again from scratch. Nowadays it is usual for many people working in research institutions to have access to a cluster of computers managed by an intelligent queue system. This tool allows to manage the load and the different users priorities efficiently. In this work we show how it is possible to continue using our old and reliable routines, written in any language, and, at the same time, to take advantage of a simple distribution technique to reduce dramatically the computational time. We will show several examples where we have applied this technique emphasizing the fact that it is not necessary to modify substantially our original programs. These examples comprise different problems in collision physics, some of them including Monte Carlo approaches, and processes in laser-atom and molecule interactions, in which it is necessary to deal with Fourier Transforms techniques.
|State||Published - Oct 2007|
- Parallel programming tools
- queue systems
- computational time optimization