Load balancing to gain performance and energy efficiency within a heterogeneous processing environment, which contains processing units with different specialized cores and memory specifications that are supposed to collaborate in a parallel way, is surveyed in the paper. Partitioned global address space (PGAS) languages are propounded as the vehicles to facilitate the communication of the heterogeneous processing collections.

With the aim of implementing dynamic load-balancing policies based on the PGAS parallel programming model, Coarray Fortran (CAF) is investigated to fulfill cooperation of remote process interactions in distributed memory systems. The overall features of CAF, with its extension, are described. The discussion proceeds with load balancing on heterogeneous nodes while considering the static and dynamic approach and their comparisons in the message-passing interface (MPI) and CAF scheduling environment. A case study concerning implementing CAF-based dynamic scheduling, for Asian option pricing, is provided. Finally, experimental results are demonstrated to evaluate the performance of different aspects of the research from multithreading, multioption, with/without multiprocessor communication, and hybrid scenarios.

The authors have composed a good technical report to explain the features and capabilities of CAF in distributed parallel systems, especially in comparison with different versions of MPI-based platforms.