Multiprocessor Real-time Systems Scheduling For Energy And Reliability Optimization

With the increasing demand for computing power of embedded systems in recent years,mul-ticore processor architecture has become more and more popular.It is an efficient parallel ar-chitecture,and its performance is significantly improved compared with single-core architecture.The multicore processor has become the mainstream chip solution for embedded systems.Chip manufacturers have introduced ulti-core processor chips,and more and more complex real-time systems have adopted the multi-core processor platforms.At the same time,designers face new research needs and challenges,including the serious energy consumption caused by the high com-plexity of the system,and the reliability problem caused by the rising rate of system failures due to manufacturing process.In addition,with the deepening of the application field and improvement of chip integration,the traditional scheduling optimization methods which purely pursue perfor-mance are no longer suitable for for the scenarios where real-time,low energy consumption and reliability assurance are considered.It is urgent to explore new scheduling strategies and optimiza-tion algorithms.Therefore,when real-time applications run on the multicore processor systems,how to schedule them to reduce the total system consumption,make the system run stably and reliably and meet the deadline constraints is still an burning issue to be solved in the research of multicore scheduling.This thesis focuses on the energy and reliability in real-time multicore processor systems.It first investigates and studies the existing scheduling techniques for energy optimization and reliability-aware energy management from different perspectives.Then,it explores the the s-trategy of energy-saving scheduling based on the execution characteristics of tasks within the applications and designs corresponding algorithms.Next,it studies the impact of existing energy management techniques on total energy consumption and reliability of the system,and comprehen-sively design the total energy and reliability optimization by rational use of energy and reliability management techniques.In addition,it explores the complete task mapping and scheduling algo-rithm to minimize system energy consumption while real-time performance and high reliability are guaranteed.The thesis includes the following three aspects:·Research on real-time multicore processor scheduling technique for energy-savings.For the scheduling problem of safety-critical time-triggered applications running on the real-time systems,this thesis first proposes to optimize the system energy consumption according to the strict and non-strict periodicity of the specific tasks within the applications.To char-acterize the strict and non-strict characteristics of the task activation,this thesis presents a novel and practical task model.Based on the task model and practical hardware,power and energy model,an integer linear programming model is proposed and solved to obtain the optimal solution of the energy-efficient real-time scheduling problem.In addition,in or-der to solve the problem of excessive complexity caused by integer linear programming,a heuristic algorithm is proposed to reduce the complexity of the solution,so that high-quality feasible solutions can be effectively obtained in an acceptable time.· Research on real-time multicore processor scheduling technique for achieving energy-savings and satisfying reliability requirements.In view of the application scenario where the relia-bility of the system must be maintained at a high level,that is,the system needs to satisfy at least the original reliability level constraint.In this thesis,two strategies are designed to ensure the high-reliable and energy-efficient scheduling.One is the ORGSEM(Origi-nal reliability guaranteed scheduling for energy minimization)which does not use DVFS to make the reliability level no longer affected by the scaling voltage/frequency.The original reliability level of system is always maintained and the system energy consumption is re-duced by adopting other energy management techniques.The other is the energy efficiency fault-tolerant scheduling(EFS).Most existing reliability-aware energy-efficient scheduling directly uses DVFS,and the use of DVFS has a negative impact on system reliability.In EFS,the introduction of PMM technique can weaken or offset the negative effects of D-VFS so as to achieve the best trade-off between system reliability requirements and total system energy consumption.Based on the above two strategies,this thesis also constructs an integer linear programming model to ensure the high reliability and minimize the total energy consumption of the system.By solving this model,the optimal voltage/frequency as-signment and scheduling solutions for energy-savings and satisfying the original reliability requirements of the system can be obtained.· Research on mapping and scheduling real-time applications on heterogeneous multicore processor systems for energy optimization and system fault-tolerance.Task mapping andscheduling are interdependent "NP-hard" problems,and the task mapping affects the energy consumption and reliability of the system.This thesis proposes an energy-efficient fault-tolerant mapping and scheduling(EFMS)framework to solve the entire problem.In the framework,the entire problem is decomposed into the mapping phase and the scheduling phase.And it is solved through two phases of iterative improvement.The task mapping solution is given by a list-based binary particle swarm optimization heuristic algorithm.The task scheduling solution is implemented by the energy-efficient fault-tolerant scheduling strategy and the corresponding algorithm.