Research On High Energy-efficient And Safety Scheduling In Heterogeneous Distributed Embedded Systems

The rapid development of very large scale integrated circuits and artificial intelligence technologies has made information technology products integrated into all aspects of our daily lives and presents significant multiprocessor,heterogeneity and networking trends.Different from the traditional distributed computing system,which is mainly used for largescale engineering scientific computing,heterogeneous distributed embedded systems are usually developed for the special needs of the industry.The system is small in scale,the coupling degree is high and the application load characteristics are obvious.The continuous development of the computing platform architecture and the needs of different application industries make the task allocation and scheduling problem still be the hotspot that needs to be studied continuously.This thesis first introduces the architecture of heterogeneous distributed embedded system,and then analyzes the research status of related fields.Next,from a view of traditional task scheduling of heterogeneous distributed systems,the thesis concerns task scheduling problem under energy constraints based on the principle of ”from simple to complex” and ”from the abstract problem to the industry application”.Then,under the background of industry applications of automotive electronics systems,which is a kind of typical heterogeneous distributed systems,the thesis focuses on and solves the problems of reliability scheduling and safety scheduling in automotive electronics systems brought by industry characters and special requirements.The main work are as follows:1)The complexity of applications and the heterogeneity of system architecture require that the computing system develop towards better energy efficiency.Energy management is a key issue in the design of heterogeneous distributed systems.In order to improve the energy efficiency in the case of limited energy consumption,this thesis studies the task scheduling optimization problem of heterogeneous distributed system with priority relation under energy constraint,and an efficient scheduling algorithm for energy consumption constrained parallel applications(ESECC)is proposed.Different from the latest algorithm which pre-allocate energy consumption beginning from the processor with lowest energy consumption,the proposed new algorithm reduces the scheduling length of parallel applications with energy constraints by assigning an average allocation value to the task in advance.The experimental results show that the proposed algorithm can obtain shorter scheduling length under the condition of satisfying the energy constraint compared with the latest algorithm in many application scenarios.2)Heterogeneous multiprocessor platform has attracted much attention in recent years due to its low power consumption,low cost and strong computing power,especially heterogeneous computing platform with reconfigurable features which has become the current research hotspot.To solve the scheduling optimization problems of reconfigurable heterogeneous distributed embedded systems under power constrains,the following steps are made.First,for the heterogeneous multiprocessor platform which combines resource delay model and reconfigurable features,descriptions of global power consumption controlling problem and flowing scheduling problems are made by employing Integer Linear Programming(ILP).On this basis,a heuristic scheduling algorithm for reconfigurable heterogeneous multiprocessor platform called PCLS(Power constrain Communication aware List Scheduling)is proposed.The experimental results shows that when the number of tasks is smaller than 35,the optimal scheduling results can be obtained within an acceptable time.The proposed PCLS heuristic scheduling algorithm can improve the scheduling performance by 10% compared with the classic DLS scheduling algorithm.For a large-scale random task test set,on different architecture test platform,using 70% of the given power constraints,the proposed PCLS only had 10% performance degradation compared to optimal scheduling,which showed good performance and adaptability.3)Functional safety aims at avoiding the unacceptable risks caused by the functional failure.For the problem of ensuring the achievement of functional safety goals in automotive electronic system,a new method that can meet the reliability goal of functional safety in automotive electronic system is proposed,which is called RGAGM(Reliability Goal Assurance Method Using Geometric Mean).The critical idea of the proposed approach relies on preallocation of reliability goal' s value to tasks according to geometric mean,through which saving more system resources.The correctness of the proposed method is proved in the thesis,and numeric simulation experiments are performed with both real automotive functional data and randomly generated distributed automotive functional data.Experimental results show that the proposed method is more effective in the insurance of reliability goals and the reduction of system resources and cost.4)The heterogeneous distributed embedded systems represented by automobile electronic system gradually evolves into mixed criticality systems.Hybrid key-level systems need to be considered on the same computing platform for multiple functions with different key levels and priorities.Mixed criticality systems need to consider the requirements of multiple tasks with different criticalities and priorities on a same computing platform.Compared with traditional task scheduling problems,the introduction of criticality attribute in mixed criticality highly increased the complexity of scheduling problem.In this thesis,FA MHEFT algorithm(Fairness and Active Multiple Heterogeneous Earliest Finish Time,FA MHEFT)was proposed,it aims at improving the performance of those tasks with high criticality.FA MHEFT algorithm uses a more positive strategy to adjust the criticality level of the system,it lowers the criticality level of the system immediately after the completion of the task that caused the promotion of system criticality level,thus giving the other functions and tasks more fair execution opportunities and optimizing the overall system scheduling length effectively.The experimental results show that the proposed algorithm can effectively reduce the DMR(Deadline Miss Ratio)of the system and reduce the scheduling length of the system as compared with the existing latest algorithms.