| 引用本文: | 李鸿飞,杜溢墨,曾熠,王磊.异构混合存储的软硬件协同数据放置策略[J].国防科技大学学报,2020,42(2):64-71.[点击复制] |
| LI Hongfei,DU Yimo,ZENG Yi,WANG Lei.Software and hardware co-design of data placement in heterogeneous hybrid store[J].Journal of National University of Defense Technology,2020,42(2):64-71[点击复制] |
|
| |
|
|
| 本文已被：浏览 282次 下载 271次 |
| 异构混合存储的软硬件协同数据放置策略 |
| 李鸿飞1,2,杜溢墨2,曾熠2,王磊2 |
|
(1. 信息工程大学 地理空间信息学院, 河南 郑州 450001;2. 中国人民解放军31008部队, 北京 100091)
|
| 摘要: |
| 分析比较当前大数据中心典型混合存储架构,针对其不能综合运用存储管理系统和存储设备优势的问题,提出软硬件协同的数据放置策略,同时考虑软件层混合存储管理系统和硬件层混合存储设备的特点,根据应用特性为数据选择合适的存储管理系统和设备。面向不同应用场景,提出运行前规划存储路径的静态放置模式和运行中规划存储路径的动态放置模式。基于存储管理系统和设备性能参数建模,采用仿真方法实现各数据放置策略,并运行实际应用中的三类负载进行实验测试,结果表明采用软硬件协同的数据放置策略相比只考虑存储管理系统和存储设备的数据放置策略,性能可以提高近30%。 |
| 关键词: 大数据 混合存储 非关系型数据库 数据放置 |
| DOI：10.11887/j.cn.202002008 |
| 投稿日期：2019-03-02 |
| 基金项目: |
|
| Software and hardware co-design of data placement in heterogeneous hybrid store |
| LI Hongfei1,2, DU Yimo2, ZENG Yi2, WANG Lei2 |
|
(1.Institute of Surveying and Mapping, Information Engineering University, Zhengzhou 450001, China;2.The PLA Unit 31008, Beijing 100091, China)
|
| Abstract: |
| The analysis of the existing typical hybrid store architectures in big data center found that they cannot fully take advantage of both hybrid storage management systems and hybrid storage devices. Thus a hardware and software co-design data placement strategy was proposed, which simultaneously considers the hybrid storage management systems at software level and the hybrid storage devices at hardware level, and figures out the trail of data placement on both storage systems and devices regarding application characteristics. Moreover, the static placement pattern before running and the dynamic placement pattern during running were proposed on the basis of different application scenes. An experiment was implemented by running three kinds of workloads on simulated data placement strategies based on the model according to the performance parameters of storage management systems and storage devices. The results show that the proposed design outperforms traditional ones that either consider storage management systems or storage devices separately by up to 30%. |
| Key words: big data hybrid store not only structured query language data placement |
|
|
|
|
|
