Mechanic: a new numerical MPI framework for the dynamical astronomy

In the field of the Solar system and planetary dynamics, extensive computational experiments became useful and well established, standard research tools. These experiments regard direct numerical integrations of complex equations of motion to study the long-term orbital evolution and stability of various N-body systems (e.g., [10]), an analysis of the resonance structure of the phase space of these systems and particular classes of solutions (like periodic and quasi-periodic orbits, equilibria), calculating dynamical maps [3] and cross-sections, modeling various types of observations of extrasolar planetary systems (radial velocities, astrometric and imaging data, eclipse and TOA timing, photometric transits) by quasi-global evolutionary algorithms (e.g.,[5]), investigating qualitative features of basic dynamical models in the framework of the general dynamical systems theory (e.g., [8]), spacecraft trajectories optimization (e.g., [4]), just to mention a few subjects of this rapidly developing branch of the dynamical astronomy. Usually, these experiments are equivalent to performing the same or very similar numerical operation or procedure on a large set of initial conditions (e.g., numerical integrations, dynamical maps) or intermediate data (e.g., when evaluating the objective function during the optimization process). In a language of computing, such calculations may be understood as standalone numerical tasks taking seconds, but also hours and days of single CPU-time. If processing of large sets of such tasks is required, one can split a given numerical experiment onto smaller parts, and distribute them over a computing pool (usually, CPU cluster or a network of workstations). This leads however to task management issues, which may be handled efficiently only by dedicated software tools. Nowadays, there are different task management systems available. Likely, the best example is the well known Condor package [13]. Within such a numerical framework, the user–supplied, stand-alone executable code performing computations is distributed over a computing pool. The input and output data of each software instance must be then handled by the host node. It requires both an efficient, possibly unified task preparation scheme, check-pointing and keeping intermediate results, as well as data storage and a post–processing (assembling the results from computing nodes). Focusing on the dynamical astronomy, we address these issues by developing a new management code, called Mechanic that is – unlike Condor and similar software systems – built on the basis of the Message Passing Interface (MPI) [7]. The MPI is highly-standardized and portable message-passing system widely used on a variety of parallel computers and CPU-clusters. We shortly present here some key features of that software (Sect. 1.), and we describe its basic usage for investigating a simple dynamical system (Section 2.). Moreover, Mechanic may become a new helper tool in a wide range of applications, particularly focusing on processing large data sets. We applied it to study the global dynamics of the ν Octantis planetary system, see our second paper in this volume. More details will be described elsewhere (Slonina et al., 2012, in preparation).