Enhanced Schedulability Analysis of Hard Real-time Systems on Power Manageable Multi-core Platforms

Abstract

Nowadays, the multi-core platforms have become the de-facto solution to cope with the rapid increase of system complexity and energy consumption. Additionally, the dynamic power management (DPM) and the dynamic voltage and frequency scaling (DVS) are two well-established techniques to adjust the trade-off between the system performance and power consumption during runtime. However, in the context of hard real-time systems the DPM and DVS have to be applied with great caution due to timing constraints. The problem of DPM/DVS based power-aware scheduling has been extensively addressed on single-core platforms. Therefore some recent studies have proposed to adapt the existing results to multi-core platforms by performing the task partition in advance. In this article, we show that this approach may not work correctly any more, if the cluster-based multi-core platforms with non-negligible DPM and DVS state switching overhead are considered. More specifically, additional delays are introduced into the task execution and thus the traditional schedulability analysis becomes insufficient. We propose a simple runtime mechanism for idle time prediction to deal with the DPM state switching overhead and two solutions to enhance the schedulability analysis by taking the DVS state switching overhead into consideration: the conservative protocol and the speed inheritance protocol. Finally, the solutions are evaluated by means of simulation.