Mapping concurrent applications on network-on-chip platforms

Abstract

Networks-on-chip have been proposed as the interconnect solution for building large systems-on-chip. Such platforms will be made of hardware cores, running concurrently to achieve high compute power and keep power consumption low. These platforms have many things in common with distributed systems. In this article we analyze the issues related to mapping concurrent applications on a networks-on-chip based platform. We show that the quality of the decisions made at different stages of mapping has a high impact on the overall system performance. The critical stages of application mapping for networks-on-chip platforms include partitioning the application into processing cores, mapping the communication between the cores, and resolving platform-dependent problems such as race conditions and multipoint-to-point communications. We investigate these problems using an MPEG4 video decoder application and we evaluate the performance of the mapped system in a simulation environment, using a SystemC networks-on-chip simulator.