Proposing a new task model towards many-core architecture

Abstract

Many-core processors are gathering attention in the areas of embedded systems due to their power-performance ratios. To utilize cores of a many-core processor in parallel, programmers build multi-task applications that use the task models provided by operating systems. However, the conventional task models cause some scalability problems when multi-task applications are executed on many-core processors. In this paper, a new task model named Partitioned Virtual Address Space (PVAS), which solves the problems, is proposed. PVAS enhances inter-task communications of multi-task applications and averts serialization of concurrent virtual memory operations. Preliminary evaluations by using micro benchmarks showed that PVAS has the potential to promote the performance of multi-task applications that run on many-core processors.