Dynamic resource management in modern multicore SoCs by exposing NoC services

Abstract

Emerging multicore chips containing tens or even hundreds of cores require modern interconnect solutions with increased programmability to support dynamic resource management. Modern embedded devices appear that employ recon-figurable architectures or application specific hardware modules, but still runtime QoS optimizations or dynamic power management require more flexibility from the underlying hardware infrastructure and corresponding middleware. This paper demonstrates a methodology to expose NoC services for adaptive management of hardware resources through a software platform based on Spidergon STNoC technology which consists of a low level driver layer and libraries accessible at user level. Thus, the system designer can exploit the runtime programmable services of a Network-on-Chip so as to provide differentiated network services to multiple independent applications. This methodology can easily be extended to any NoC technology. Spidergon STNoC allows to design customized topologies through the iNoC GUI tool, which is extended to generate the appropriate driver for the Linux kernel. Moreover, an integrated C API allows the developer to capture application specific requirements and dynamically adjust QoS settings of the NoC. In this paper we improve the design methodology to facilitate dynamic management of SoC resources with the aid of appropriate driver and library extensions; we present tools that offer extreme flexibility and real examples of software applications that can exploit the NoC configurability, running on both typical Linux and Android environments.