Combined Dynamic And Static Vertical Power-Efficient Strategy

The energy problem in embedded system emerges with its mobility. Because of the popularity of embedded application, system energy grows rapidly. Battery technique cannot forward as fast as energy. The energy problem becomes more and more severe.Hardware method and software method are exploited for saving power. Software method includes static and dynamic method. Static method scans the code at compile-time and finds the segment that could be optimized for target platform. Dynamic Voltage Scaling and Dynamic Power Management are two mostly used methods. Because the power of circuit has linear square relationship with the input Voltage, lowering it can gain obvious effect. Dynamic Power Management adjusts the next state of device by current work-load and state. There are three main algorithms of Dynamic Power Management: timeout, prediction and random process. Embedded storage devices do not gain much attention on their power saving problem. With the increase of data processing need, the sensitivity of storage device power grows up.This paper presents the vertical power-efficient model for disk which combines the static and dynamic methods. This model improves the code to be executed and the device. In order to take full advantage of code and device, it incorporates these two kinds of methods.The static code restructuring methods focus on large-size array access in nested loops. After discussing access pattern, data locality in cache and memory, presents three static optimizing methods: loop fusion, loop fission, loop tiling, implemented these methods on simulating platform with satisfying result. We exploited a new dynamic power-saving method, called Request Accumulation Threshold Scheduling, which is based on timeout and prediction methods. Compared to the ordinary timeout method, it can reduce at least 20% energy.