Reliability-driven Task Scheduling In Real-time Systems

With the continuous advances in IC manufacturing technologies,the size of modern microprocessors has been significantly reduced and the performance has been highly im-proved.This greatly promotes the wide application of real-time embedded systems in key areas such as environmental monitoring,space exploration,industrial process control,flight management and control,financial administration,automotive electronics,smart appliances and medical devices.At the same time,due to the ever-increasing number of transistors on a chip and high integration level,microprocessors are more vulnerable to electromagnet-ic interference-,instantaneous voltage disturbance-,and high energy particle strike-induced transient faults.This could degrade system reliability.However,high reliability is a nec-essary requirement to ensure the successful operation of the aforementioned applications.Therefore,designing a reliable real-time embedded system becomes an inevitable trend.Reliability-driven real-time task scheduling mechanism is an important means to guar-antee the high reliability of the system,thus has always been a hot topic of real-time em-bedded systems.The mechanism improves the system reliability by using techniques such as rollback recovery,redundancy,and frequency boost that can either tolerate the occurred transient faults or lower the fault arrival rate.However,these techniques have the side ef-fects of elevated energy consumption,increased makespan(thus reduced throughput),and higher temperature(thus shortened device lifetime).Traditional reliability-driven real-time task scheduling mechanisms often neglect these problems.This work aims to jointly tack-le these problems and design novel reliability-driven real-time task scheduling algorithms that optimize system energy consumption,makespan,and device lifetime while ensuring the reliability and real-time requirements.Specifically,1.This paper first addressed the problem of energy minimization under the reliability and peak temperature constraints,and proposed reliability-driven energy-efficient real-time task scheduling mechanisms for single processor and multiprocessor systems.For the studied problem of single processor systems,this paper introduced a fault adaption variable and exploited the checkpointing technique to meet the reliability constraint,proposed an energy efficiency factor that determines the operating frequency of tasks to reduce the energy consumption,and designed an improved thermal-aware task se-quencing technique that lowers processor temperature to satisfy the peak temperature constraint.For the studied problem of multiprocessor systems,this paper also used the fault adaption variable and checkpointing to ensure the reliability constraint,em-ployed the thermal-aware task sequencing technique to ensure the peak temperature constraint,and presented a dynamic energy-optimum task allocation scheme to reduce the energy consumption.In addition,heuristic algorithms are developed based on the scheme to minimize the system energy consumption under all the constraints.2.This paper then addressed the problem of makespan minimization under the reliabil-ity and peak temperature constraints,and proposed the reliability-driven makespan-optimum real-time task scheduling mechanism for multiprocessor systems.The op-timization problem is modeled and solved using mixed-integer linear programming(MILP).Considering the limitation of MILP,this paper explored and analyzed the effects of task assignment on makespan,reliability,and temperature.Based on the analysis,a heuristic task assignment and scheduling algorithm that utilizes the sched-ule length balanced task allocation,reliability-aware redundancy,and thermal-aware task sequencing and frequency scaling is proposed to minimize the makespan while satisfying all the constraints.3.This paper finally addressed the problem of jointly optimizing system reliability and lifetime,and proposed reliability-driven lifetime-aware real-time task scheduling mech-anisms for single processor and multiprocessor systems.For the studied problem of single processor systems,this paper presented a novel analytical method to calculate the mean time to failure(MTTF)due to transient faults.Using this new method,the joint optimization of system reliability and lifetime is transformed into the optimiza-tion of system availability,and a framework is designed to optimize the availability of single processor systems.For the studied problem of multiprocessor systems,this paper also transformed the joint optimization of system reliability and lifetime into the optimization of system availability,and proposed a scheme to optimize the availability of multiprocessor systems.