Achieving Thermal-Resiliency for Multicore Hard-Real-Time Systems

Abstract

Multicore processor based system designs are increasingly utilized as the processing platform for complex hard-real-time and embedded applications. These real-time systems need to operate under various physical and design constraints. Much research has focused on thermal-aware real-time systems designs. However, no results exist to investigate the resource allocation and the system degradation under external thermal constraints in a predictable manner. This paper proposes a control-theoretic framework to ensure hard-real-time deadlines on a multiprocessor platform in a dynamic thermal environment. We use real-time performance modes to permit the system to adapt to changing conditions. Also, we show how the system designer can use our framework to allocate asymmetric processing resources upon a multicore CPU and still maintain thermal constraints. We develop analysis for determining what modes the system can support for a given external thermal condition. Our system design extends the derivation of thermal-resiliency (originally proposed for uniprocessor systems) to multicore systems and determines the limitations of external thermal stress that any hard-real-time performance mode can withstand. Simulations and physical test bed results show that our algorithm predicts how a system will gracefully and predictably degrade under external thermal stress.