Event-driven Embedded Pma And Scheduling Algorithm

With the development of embedded computing technology, embedded chips have become more and more powerful on one side but are more and more restricted to the battery power supply technology whose development lags far behind. The continual working time, manufacture and run cost, encapsulation bulk, reliability, life circle, and the use of high performance components of the embedded systems are hence highly affected and limited by embedded system power consumption. Power consumption has become one of the most crucial constraints in embedded system design. Many new hardware power management feasures made software power management the most important means in controlling and optimizing system power consumption. It is also among the hottest topics in embedded system research these years. This thesis systematically researchs system power management technologies and mainly contains the following two fields.First, propose a dynamic voltage and frequency scaling (DVFS) policy based on task decomposition: Traditional software DVFS policies ignore the fact that memory system is usually the bottle neck of a system. They tend to simply believe that system performance is linear to CPU core frequency. This results in poor effects when we put them into real use. Targeting at this, this thesis proposes a new DVFS policy using task decomposition. It divides system running time into many equal intervals, and analyzes tasks workload composition at the end of an interval. It computes the dependence degree of a system on its off-chip memory. It also does analysis on system power classification in order to do proper system power estimation. To garanntee system performance, it uses its own performance loss standard to choose eligible CPU frequceny points. Finally the one which brings the minimum power consumption is chosen as the CPU frequency for the next interval. The policy can achive both goals of optimizing power and guarantee performance.This policy is implemented on Intel Xscale based platform. Using PXA270 PMU, task decomposition computing and DVFS policy is realized. The results is given and analyzed. Second, propose an event driven power management architecture (PMA): An example PMA is disscused together with its core idea and realization on PXA 270 board. Thesis gives out judgement for it and proposes an event driven PMA. This PMA uses the thought of separting mechanism and policy. It supports DVFS, dynamic device power management, system low power mode management. It also provides policy framework for power management policies making. The exact policy can be configured or implemented according to different needs making this PMA very flexible. Finally this PMA is implemented on Intel Xscale based platform using PXA270 processor.