Integration of Heterogeneous and Non-dedicated Environments for R

Parallel computing is becoming essential for nowadays data analysis in several disciplines. In order to profit from parallel processing of experimental data, specialized skills, software tools and suitable computing resources are required. Desktop grids and volunteer-based systems have proved themselves as powerful options where distributed idle resources from heterogeneous computers are aggregated to build powerful met computers. Software solutions are required to automate and assist the process of transformation and adaptation of current and new applications to run in these environments. Finally, it is desirable, for the same tool, to provide an efficient solution to orchestrate the execution of these programs using a diversity of dynamic environments. In this paper we describe an implementation of an integrated solution for the R language which allows the transformation and execution of parallel loops in heterogeneous and non-dedicated environments. The results obtained allow us to prove the feasibility of our proposal. Furthermore, several issues that tools like this must consider to improve their performance when integrating heterogeneous systems are described.