Network-on-Chip Protection Switching Techniques for Dependable Task Migration on an Open Source MPSoC platform

Inhalt:
Transient and permanent faults increasingly affect the behavior of future multiprocessor system-on-chip. Thermal hotspots for example can induce stress which may either lead to impairments of signal integrity or even to physical damages in form of aging. In this paper we present our work-in-progress on dependable task migration to mitigate these issues. Triggered by a thermal management system, tasks are migrated to colder processor cores together with a reliable switch-over of the associated communication channels. For this we apply appropriate protection switching mechanisms which allow a fast migration with short application downtime. Assessing new concepts for future many-core systems on a physical hardware platforms is challenging due to the complexity and the variety of hardware and software components, especially if specific hardware IP is involved. To provide researchers a good starting point we propose an open source MPSoC platform which we introduce in this paper. The platform integrates all necessary elements to build simulation- or FPGA-based prototypes to evaluate and validate concepts. We use the proposed platform to investigate our task migration concept using system-level simulation which is also part of the presented open source MPSoC platform.