1 Introduction and motivations

In recent years, personal computers (PCs) became very efficient computational devices. Processors of the last generation PCs can rival nowadays with those of the more traditional workstations, with the additional advantage of being cheaper. As a consequence, clusters of PCs with their processors connected in parallel turn out to be a relatively economical way to reach a very high performance of computation. (In particular, this becomes the only accessible tool in those places with limited access to computational resources). Obviously, such a parallel arrangement is not suitable to the running of a classic sequential code; thus, from the appearance of clusters of PCs has arisen the necessity of developing strategies for programming in parallel.

Besides this, it is well known that discretization problems arise when simulating a large system such as a typical galaxy ( $\simeq10^{11}$ stars) with sets of 10⁴ or even 10⁵ particles, see for example Hernquist & Barnes ([1990]): whereas a typical star moves within a smooth potential in a real galaxy, a typical particle suffers multiple collisions during the simulation, thus resulting in spurious relaxation effects. Thus, a large number of particles is needed in order to properly simulate such systems. As the number of particles grows, so does the computational time involved, and fast machines and efficient algorithms become vital. Clusters of PCs and parallel algorithms come to satisfy these needs.

The paper is organized as follows: Sect. 2 describes the sequential features of the code, whereas Sect. 3 deals with those aspects concerning parallel programming. Some tests and simulations are presented in Sect. 4. The conclusions are considered in Sect. 5.