对象存储系统中数据分布及重构算法研究-计算机系统结构专业毕业论文.docx

Abstract*
Computer storage system research is important to puter science, next puter storage architecture——object storage is a hot point to puter science. Object storage has many advantages such as high work oriented and data management intelligence makes it fit to use in the large work storage and intelligence storage system.
Object storage system uses meta data technology to manage the data efficiently and uses data stripe technology to make the data accessing easily, but this two mechanism bring some problem to the scalability of object storage system. Through analyze of the mechanism of object storage system and the problem of scalability, we find contravention between the mechanism of object storage system and the scalability, and consistency of metadata is another vital point to the object storage system. In order to keep the advantages in the object storage system and make the scalability of the object storage system, we use a new data placement ——D/H data placement to replace the origin mechanism which just use the simple round-robin method, and have designed a simulate system Lustre-Lite to make experiment.
The test indicate that using the D/H data placement method makes object storage system more scalability and adds function of data anization online. Analyze of experiment brings us that object storage D/H data placement algorithm in data anization makes space balanced、dynamic load least and load balanced, it takes prominence benefit to replace the origin round-robin algorithm.
Keywords: Object Storage, System Scalability, Data Distribution, Data anization, Quality of service
* The research is supported by National Basic Research Program “973” program (2004CB318203) GI “Intelligent storage system of WAN”(CNGI-04-5-1D).
目 录
摘 要............................................................................................................... I
Abstract........................................................................................................