低能耗存储系统的设计与实现

发布时间：2018-12-17 06:35

【摘要】：为了满足呈爆炸式增长的信息存储、处理、传输的需求，大规模数据中心应运而生。在大规模数据中心里，根据经典的80/20原理，将数据都存储在高性能设备上是不经济的，为了实现资源的优化配置，于是就有了分级存储概念。研究表明数据中心的服务器很少处于满负荷状态，由于服务器消耗的能量与服务器的利用率并不成比例，空转的服务器（空闲的CPU、空转的硬盘等）浪费了大量的能量。设计一种低能耗存储系统就十分必要。通过对现有的x86服务器在性能、能耗方面进行分析，设计了一种低能耗存储系统——基于嵌入式平台的分级存储系统。首先，介绍了低能耗存储系统的应用场景，讨论了在系统实现中涉及到的关键技术，分析了在实现基于数据访问热度和存储资源性能的在线数据迁移机制时可能引发的相关问题。其次，对原型系统进行设计和实现，构建了由应用服务器、存储服务器和管理服务器三个部分组成的系统。系统包含系统监控、存储管理、数据管理、智能管理和用户界面五个模块，系统监控模块主要收集用户数据访问热度信息和存储服务器的负载信息；智能管理模块根据这些信息调度数据进行迁移，并对服务器有选择地进行开启和关机，在实现基于用户访问频度的分级存储机制的同时，达到节能降耗的目的。最后通过实验验证了该系统的功能，对系统性能进行了评测，分析了通过数据迁移对系统进行调度系统能耗的变化。测试结果表明，，它能够依据数据的实时热度和服务器的负载，对数据进行自适应迁移；再通过对服务器启停进行动态调度，可以在实现优化资源配置的同时，将系统能耗降低25%。
[Abstract]:In order to meet the demand of explosive information storage, processing and transmission, large-scale data center emerged as the times require. In large scale data centers, it is not economical to store data on high performance devices according to the classical 80 / 20 principle. In order to optimize the allocation of resources, there is the concept of hierarchical storage. The research shows that the server in the data center is rarely in the full load state. Because the energy consumed by the server is not proportional to the utilization rate of the server, the idle server (idle CPU, idling hard disk, etc.) wastes a lot of energy. It is necessary to design a low energy storage system. By analyzing the performance and energy consumption of the existing x86 server, a low energy storage system-hierarchical storage system based on embedded platform is designed. Firstly, the application scenario of low energy consumption storage system is introduced, and the key technologies involved in the system implementation are discussed. This paper analyzes the related problems in the implementation of online data migration mechanism based on data access heat and storage resource performance. Secondly, the prototype system is designed and implemented, and the system is composed of application server, storage server and management server. The system consists of five modules: system monitoring, storage management, data management, intelligent management and user interface. The system monitoring module mainly collects user data access heat information and storage server load information. The intelligent management module migrates according to these information scheduling data and selectively opens and shuts down the server so as to realize the hierarchical storage mechanism based on user access frequency and achieve the purpose of saving energy and reducing consumption. Finally, the function of the system is verified by experiments, the performance of the system is evaluated, and the energy consumption of the scheduling system is analyzed through data migration. The test results show that it can transfer the data adaptively according to the real-time heat of the data and the load of the server. Through dynamic scheduling of the server, the system energy consumption can be reduced by 25% while optimizing the allocation of resources.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP333

【参考文献】