基于能耗数据的嵌入式系统内存压缩技术的研究与应用

发布时间：2018-09-13 06:57

【摘要】：能源是是人类社会发展必不可少的物质基础，但能源紧缺的态势日趋严峻，因此社会和国家政策积极倡导节能减排。华南理工大学能耗监管平台顺应国家节能政策这一大趋势，按照《高等学校校园建筑节能监管系统建设技术导则》进行建设，完成了南北校区总计129栋近100万平方米各类建筑能耗计量工程，实现了能耗智能实时监控、能耗统计等功能。能耗监管平台物理架构中的数据中转站主要负责对末端水电冷数据采集器组成的物联网（Internet Of Things, IOT）进行数据采集、数据缓存、数据发布等，运行于传统的PC服务器上，而PC服务器硬件成本高，占空间大，能耗大，故研究尝试将其移植到成本低廉省空间能耗小的Linux嵌入式设备中。但由于数据中转站要缓存的数据量大以及程序中多线程的资源消耗多，嵌入式设备的内存资源往往不够，经常发生内存交换，即磁盘I/O，导致内存的访问效率低下，制约了整个系统的性能，使得数据中转站中转数据的响应变得非常慢。为缓解这个严重问题，提高嵌入设备的系统性能，本文引入了内存压缩的机制，设计和实现了一个可动态装载卸载的内存压缩系统。内存压缩系统的主要思想是，在内存中预分配一块区域作为虚拟交换区，将换出的页以压缩的形式存放在该虚拟交换区。当发生请求调页时，系统从虚拟交换区中找到该页，，将其解压缩后换入内存，从而避免了页的换入换出这两个硬盘访问的低速过程。该系统是以块设备驱动程序的形式来实现的，其好处是无需修改内核的源码，而且模块可以在无需重新启动系统的情况下动态的装载和卸载。此外，根据能耗数据局部和全局重复性都很高并且以字符串形式存储的特点，本文设计出了一种基于字典的自适应无损压缩算法MLZ（Mixed LZ）。它是一种基于LZ77和LZW的混合型改进算法，以LZW算法为主LZ77算法为辅，利用了它们的互补特性，减少了扫描时间以及滑动窗口中字符串比较匹配时间，具有更好的全局与局部自适应性、更高的压缩率。最后本文对该嵌入式内存压缩系统进行了性能测试，基于华南理工大学能耗监管平台，将华南理工大学所有的能耗数据点作为测试工作集，在相同的系统配置下，在本地调用webservice接口分别从使用和未使用内存压缩系统的嵌入式设备以及未使用内存压缩系统的PC机上的数据中转站中获取能耗数据点的值，并计算花费的总时间，以此判断使用之后内存压缩机制是否提升了系统性能以及是否满足实际部署需求。实验结果表明，使用了本文实现的内存压缩系统达到了预期目标。
[Abstract]:Energy is an indispensable material foundation for the development of human society, but the situation of energy shortage is becoming more and more serious. Therefore, social and national policies actively advocate energy conservation and emission reduction. Construction, completed the North-South campus of a total of 129 buildings nearly 1 million square meters of various building energy consumption measurement projects, to achieve intelligent real-time monitoring of energy consumption, energy consumption statistics and other functions.
The data transfer station in the physical architecture of the energy consumption monitoring platform is mainly responsible for data acquisition, data caching and data publishing of the Internet of Things (IOT), which is composed of the terminal water and electricity cooled data collector. It runs on the traditional PC server. The PC server has high hardware cost, large space occupation and large energy consumption, so the research tries to move it. Because of the large amount of data to be cached by the data transfer station and the consumption of multi-thread resources in the program, the memory resources of the embedded devices are often insufficient. Memory exchange, i.e. disk I/O, often occurs, which leads to the inefficiency of memory access and restricts the performance of the whole system. In order to alleviate this serious problem and improve the system performance of embedded devices, this paper introduces the mechanism of memory compression, designs and implements a memory compression system which can be loaded and unloaded dynamically.
The main idea of the memory compression system is to pre-allocate an area in memory as a virtual swap area and store the pages in the virtual swap area in the form of compression. The system is implemented in the form of a block device driver, which has the advantage of not modifying the source code of the kernel, and that modules can be loaded and unloaded dynamically without restarting the system. In this paper, a dictionary-based adaptive lossless compression algorithm MLZ (Mixed LZ) is designed. It is a hybrid improved algorithm based on LZ77 and LZW, supplemented by LZW algorithm as the main LZ77 algorithm, using their complementary characteristics, it reduces the scanning time and the matching time of string comparison in sliding window, and has better global and global performance. Self adaptability and higher compression ratio.
Finally, the performance of the embedded memory compression system is tested. Based on the energy consumption monitoring platform of South China University of Technology, all the energy consumption data points of South China University of Technology are taken as test worksets. Under the same system configuration, the embedded devices of using and not using memory compression system are invoked by local web service interface, respectively. The experimental results show that the memory compression system implemented in this paper achieves the performance of the system and meets the actual deployment requirements. Expected target.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP368.1

【参考文献】