Energy Efficient Multicore Scheduling Algorithms for Real Time Systems

Abstract

With the advancement of technology and ever increasing demand of portable, scalable and sophisticated embedded systems, managing energy consumption to prolong the battery life of embedded devices has become a big challenge. With the advent of multi-core processors in the embedded market, reducing the energy consumption is becoming increasingly important for multi-core processors as well. Modern multi-core processors consume two types of energy, viz., dynamic and static energy. Dynamic energy is consumed due to switching activity whereas static energy is consumed due to increase in leakage current. These processors have capability to dynamically lower the supply voltage that reduces dynamic energy consumption. However, reducing supply voltage increases gate delay which requires one to lower operating frequency. As a consequence, the tasks take more time to execute. In this thesis, we have focused on real time embedded systems that execute hard and soft real time tasks. The major challenge for these systems is to optimize energy consumption using dynamic voltage and frequency scaling (DVFS) without missing the timing constraints of the hard real time tasks and responsiveness of the soft real time tasks. The energy saving achieved by DVFS is severely limited with the dramatic increase in leakage power consumption. Therefore, to minimize the overall energy consumption, there is a need to optimize dynamic as well as static energy consumption. This thesis addresses the issue of overall energy optimization in real time embedded systems at the operating system level using efficient real time task scheduling algorithms. The proposed energy efficient scheduling algorithms, Energy Efficient Dynamic Voltage and Frequency Scaling (EEDVFS) and Energy Efficient Uni-Core Scheduler (EE-UCS) optimize dynamic energy consumption of uniprocessor. Another proposed energy efficient scheduling algorithm, Multi-Core Scheduler (MCS) optimizes dynamic energy consumption of homogeneous multi-core processors. These algorithms are capable