Tolerisanje grešaka i energetska efikasnost kod sistema za rad u realnom vremenu sa vremenskom redundansom

Abstract

The concept of real-time systems (RTSs) is presented in the computer science for decades. During that period, the RTSs have evolved from special purpose microcomputer systems for industrial application to various forms of embedded system that are deeply ingrained in wide segments of daily life. The new application domains pose new design requirements and goals to RTSs, which are now often required to provide both fault tolerance and energy efficiency in addition to their main objective to compute and deliver correct results within a specified period of time. There is a fundamental tradeoff between these two additional requirements because fault tolerance techniques use slack time to improve reliability while low energy consumption techniques exploits slack time to increase energy efficiency. The central problem considered in the dissertation is how to optimally distribute the slack time between these techniques. Dynamic voltage scaling (DVS) is known as one of the most effective low-energy technique for RTSs. However, most existing DVS techniques only focus on minimizing energy consumption without taking the fault-tolerant capability of RTSs into account. In order to solve specify problem in this dissertation, a new heuristic-based fault-tolerant dynamic voltage and frequency scaling (FT-DVFS) algorithm is developed. The goal of the proposed algorithm is to minimize the amount of energy consumed by a real-time system under fault tolerance constraints while guaranteeing that all real-time tasks can complete successfully before their deadlines. Basically, the FT-DVFS is a DVS algorithm with integrated response time analysis (RTA) to check both the schedulability and the fault tolerant constraints of real-time task sets. The performances of FT-DVFS algorithm are evaluated by simulation in a custom build simulator. The simulation results are analyzed from three different points of view: the schedulability, the energy consumption, and the fault tolerance. The simulation results show that the proposed algorithm saves a significant amount of energy even with only two frequency/voltage levels, and the savings further increases with the increase of the number of frequency levels. Also, the simulations show that the reduction in power consumption, which can be achieved with FT-DVFS algorithm decreases with the increase of the processor utilization factor (i.e. processor spare time). The simulation results from the fault tolerant point of view show that the higher level of fault tolerance can only be attained through sacrificing a part of savings in power consumption, and vice versa. The proposed heuristic FT-DVFS algorithm is compared with the optimal DVS algorithm. The simulation analysis show that FT-DVFS algorithm achieves near-optimal solutions in very short computation time even for large task sets.