Software platform for hybrid resource management of a many-core accelerator for multimedia applications

Abstract

As incessant demand of higher computing capability makes a many-core accelerator become a major computing resource in a System-on-Chip, a variety of many-core architectures and resource management techniques have been proposed recently. They usually assume a specific hardware architecture and a specific resource management scheme. In this paper, we propose a generic software platform that implements a hybrid resource management technique, targeting for a wide range of many-core architectures. To evaluate the system performance more accurately before SoC fabrication, we run it on a virtual prototyping system. The actual implementation enables us to investigate the overheads involved in the propose software platform. Preliminary experimental results confirm that the proposed software platform adapts to the dynamic workload variation effectively by dynamic mapping of tasks and tolerate unexpected core failures by check-pointing. We address our perspective on future research issues to make the generic software platform a reality.