Towards Actor-oriented Programming on PGAS-based Multicore Architectures

Abstract:
Actor-based programming is widely used in concurrent systems to express dependencies between tasks and exploit task-level and pipeline parallelism. Current and emerging multicore architectures induce big challenges in the practical applicability of actors in such systems due to NoC-based distributed memory hierarchies and heterogeneity of processing resources. PGAS-based parallel programming languages tackle these problems and provide concepts for shared memory parallelism and message passing between several heterogeneous compute tiles in a common language. In this position paper, we present a first framework for actor-based programming concepts embedded into the PGAS-based programming language X10 in order to exploit its programmability and portability in the context of heterogeneous multicore architectures. We will show first results by applying the proposed concepts to a complex computer vision task chain written in X10.