TOWARDS THE EVOLUTION OF LEGACY APPLICATIONS TO MULTICORE SYSTEMS Experiences Parallelizing R

Current innovations in processor performance, focused to keep the growth rate of the last years, are mainly based on providing several processing units within the same chip. With new underlying multicore processors, traditional sequential applications have to be adapted with parallel programming techniques to take advantage of the new processing capabilities. There exists a great variety of libraries, middlewares, and frameworks to assist the parallelization of such applications. However, in many cases, specially with classical scientific applications, due to several limitations ranging from technical incompatibilities to simply lack of knowledge, this evolution cannot always be achieved. We here present our experiences providing an alternative for two situations where former contributions could not provide a satisfactory solution to our needs: adapting a mature non-thread-safe C coded application, the R language interpreter, and providing support for the automatic parallelization of R scripts in multicore systems.