Research On Low-Power Techniques For Soft Real-Time Embedded Multi-Core Processor System

High-power and high-heat processor has always been bottlenecks of computer system's design. For their own characteristics of applications, the system performance and power consumption of embedded system have more stringent requirements. Therefore, the study of low-power techniques is of utmost importance to research and development of embedded system. Especially in recent years, rapidly increasing demands of the soft real-time system coming with the extensive application of multimedia techniques, and the rapid development of multi-core processor techniques provide a new and effective way of low-power issues. This thesis researches the design, implementation and application of low-power algorithms for soft real-time embedded multi-core processor system, based on the design of a low-power soft real-time embedded digital media system.This thesis firstly presents the overall framework, spatial distribution maps and functional partitions of a low-power soft real-time embedded digital media system, and analyses the composition of the various subsystems' framework and functions in detail. Then, takes the embedded distributed sub-server of the digital media distributed system as an important studying object, and discusses low-power techniques on its operating system layer and application one.In the operating system layer, taking into account the existing VAP_M's characteristic, which can provide optimal solution but its research environment is too idealistic, and the various inter-processor communication costs, presents the improving static low-power algorithm(VAP_SA, Voltage Assignment with Probability_StAtic) used for resolving VAP(Voltage Assignment with Probability) problem, and inspired from online greedy scheduling model which uses (m,k)-firm constraint to solve the problem of low-power, combined with the improved static low-power algorithm, gives the dynamic low-power algorithm(VAP_DY, Voltage Assignment with Probability_DYnamic) for VAP problem, which scales voltages online. Theoretical analysis and experimental results show that, compared to the existing low-power algorithm, improved static method and presented dynamic algorithm can take full advantage of uncertainty of tasks' execution times, scale system voltage at right time and effec- tively lower system power consumption while guaranteeing the time constraint and given completion rate.In the application level, with the present situation that widely used video streaming media techniques make the requirements to control rate of non-TCP flow to ensure network's performance become very necessary, improves the existing TCP Reno throughput model, and gives the new low-power dynamic algorithm for soft real-time multimedia systems' applications—Improved TCP-friendly low-power algorithm(LPTF, Low-Power TCP-Friendly algorithm) to meet network performance's requirements of the soft real-time embedded digital media system, and achieve the purpose of rightly reducing system's power consumption according to current performance of network, applied improved TCP Reno throughput model, combined with dynamic voltage scaling techniques and current TCP-friendly algorithm. Experimental analysis show that improved model can predict current throughput of network more precisely and the new low-power TCP-friendly algorithm can reduce system's power consumption effectively while guaranteeing network's transmission.