Research On Energy Efficient And Reliable Optimization Algorithms Of Embedded Real-time Systems

With the development of computing technology, multiple or many cores networked embedded computing, low power green computing and highly dependable computing all bring new design and optimization problem for embedded real-time systems. Embedded systems have been used in more and more applications and the chips integration level are increasing. The co-optimization of real-time, energy and reliability of embedded systems are particularly important, and there is urgent need of new research ideas and methods.This thesis summarizes the design method of energy saving, fault-tolerance and trade-off of energy/reliability for co-synthesis of uniprocessor, multi-processor(core) and Network-on-Chip(NoC) systems. Several key design problems are identified and solved for embedded applications with different platforms, design needs and performance constraints. These problems include:(1) hardware-software partitioning and multiple attributes evaluation of embedded system,(2) fault-tolerant scheduling and tasks voltage assignment of dual-processor,(3) IP core mapping and links voltage assignment of NoC.The main works and contributions are as follows:(1) A multi-attributes decision making based algorithm is proposed for multi-objective hardware-software partitioning of embedded systems, under the area, cost, power, real-time and reliability constraints. Given the task flow graph and a library of software or hardware component, reliability is transformed to an additive attribute of cost type firstly, and a constrained hardware-software partitioning Pareto set is obtained by a modified shortest path algorithm. Then the multi-objective optimized partitioning result is obtained by evaluation and ranking of the candidate partitioning scheme, using TOPSIS(Technique for Order Preference by Similarly to Ideal Solution) with combined weight.(2) A reliability optimization algorithm based on Tabu Search, is proposed for energy-constrained fault-tolerant scheduling and task voltage assignment problem of dual-processor real-time embedded system. Given the independent real-time task sets and two voltage-scalable processors, EDF(Earliest Deadline First) is used to schedule on each processor based on modified primary-backup model. Considered the impact of voltage on faults, the backup number of every task on each processor and the voltages of each task instance are optimized to maximize the system reliability under the reliability constraints of safety-critical tasks.(3) A nested energy/reliability co-optimization algorithm is proposed for energyand reliability-aware No C mapping and links voltage assignment problem. Given the IP core communication task graph and links voltage scalable NoC platform, Genetic Algorithm is applied to NoC mapping and a multiple performance gradient driven heuristic is proposed to assign the voltages of each task on mapped links. The proposed approach, and can reduce the communication energy efficiently under the NoC bandwidth constraints.With the rapid development and wide application of embedded systems, smart electronic products and mobile Internet are in the stage of initial development. There are many new and open problems of energy saving and reliability optimization of embedded systems. The design methods and energy optimization algorithms presented in this thesis may provide some new techniques and ideas to energy/reliability trade-off of embedded systems.