Research On Several Issues Of Embedded Systems Optimization

To achieve a high performance, secure and low energy consumption embedded systems has always been an important goal for designers to pursue. With the rapid development of semiconductor technology and computer architecture, the performance of embedded hardware platforms has been promoted substantially. However, to improve the performance of embedded application it is necessary to optimize the performance of software according to new architecture features of embedded platforms. With the development of embedded network application, embedded systems also face huge security threats. How to get a trade-off between security, real-time and energy overhead is a challenging problem for constructing security embedded systems. Furthermore, with the development of mobile application and tiny embedded systems, reducing energy consumption of embedded systems has become to be a hot research topic of embedded systems field.The purpose of this dissertation is to study some concrete problems in embedded systems optimization. First, based on how to optimize on-chip memory's efficiency, we conducted some researches on code optimization technique for some novel processor and memory architectures. Then, we studied security optimization problems for embedded real-time systems. In the end, we conducted research on how to enhance the security strength of energy-constrained embedded systems based on wireless sensor nodes. The main contributions of this dissertation are:(1) We studied variable partitioning and task scheduling problems for the MPSoC architecture. First, we designed a loop data flow graph model named LDFG, which can describe memory reference of each task and task dependency between different loop iterations. Then, we proposed a very efficient variable partitioning algorithm named HAFF. Base on the concept of potential critical path, we proposed another algorithm named GVP, by which we can get good variable partitioning for overall performance. Based on rotation scheduling, we proposed a loop pipeline scheduling algorithm named RSVP. Using RSVP can make full use of parallelism features of MPSoC and its memory subsystem, so that the overall performance of application can be improved drastically. (2) We studied the code optimization problem for multi-module memory. First, we formulate variable assignment, instruction scheduling and energy mode setting problems in a unified ILP framework named PEOS. The significant advantage of PEOS is that it can achieve the optimal performance or energy consumption solution for code optimization problems. During the research process of PEOS, we proposed two common theories to model nonlinear conditions"c=0 iff r=0"and"c=0 iff r=1"by linear conditions. To improve the practicality of PEOS, we proposed two acceleration methods for model solving: 1) schedule window compact, 2) VNS-ILP which is an approximate method based on variable neighborhood search. The former can guarantee the optimality; the later can get near optimal solution in a very short computation time.(3) We studied how to generate optimal security strategy for embedded real-time systems. First, we proposed a security-related task graph model named STG,which can express security requirements of each task,task dependency and real-time constraint. In the second place, we quantitatively evaluated performances and security strength of common cryptographies. Based on these results, we proposed a quantitative model to evaluate the system overall security strength. In the end, we proposed an ILP based method named ILPOS, which can achieve the optimal security strategy for the specific application while guaranteeing real-time schedulability. For an application can be represented by path or tree structured STG, we proposed two dynamic programming based algorithms, which can generate the optimal security strategy in polynomial time.(4) We took sensor network nodes as platform, and studied how to apply security algorithms in energy-constrained embedded systems by experimental method. First, we designed a circuit for micro power measurement, and measured energy consumptions of various security algorithms on CrossBow and Ember sensor nodes. Then, we proposed some code optimization methods to reduce energy consumption of security algorithms. In the end, based on analysis of measurement results, we proposed some principles for using security algorithms in sensor network nodes, such as cryptography selection, parameter configuration and so on. These principles can be also applied to other energy-constrained embedded systems.