中文摘要：

随着数据存储规模的飞速增长,降低存储系统的总拥有成本,提高数据访问性能成为构建海量存储系统的关键.设计并实现了一个海量数据分级存储系统TH-TS（Tsinghua Tiered Storage）,由多级存储设备构成一体化的数据存储环境.该系统提出了Cute Mig数据迁移方法：采用基于升级成本和升级收益的升级迁移策略和基于剩余空间的文件自适应降级选择策略,解决了传统on-demand迁移方法中迁移数据量大、访问性能不佳的问题.评测结果表明,TH-TS采用Cute Mig迁移方法的系统平均I/O响应时间比传统的LRU和GreedyDualSize方法分别降低了10%和39%左右,数据升级迁移量分别降低了32%和59%左右,降级迁移量分别降低了47%和66%左右.

英文摘要：

With the rapid growth of storage scale,the key issues in building massive storage systems are to reduce the total cost of ownership（TOC）and to improve system performance.In this paper,the design and implementation of a tiered storage system,called TH-TS（Tsinghua Tiered Storage）,are presented.In TH-TS,we design an approach for migrating files efficiently in tiered storage system（CuteMig）,which promotes the files based on their promotion costs and benefits with the incremental scanning technique to acquire files＇ access information and adaptively demotes files based on the free space of a high-end storage tier to ensure that the high-end storage system will have free space.It eliminates file replacement before promotion,and when files are promoted,it reserves their replicas in the low-end storage system.This approach solves the shortcomings of the traditional on-demand promotions and replacement demotion approaches,resulting in improved system performance and less amount of migrated data.The evaluation results reveal that TH-TS could effectively migrate files among different tiers,and reduce the average response time by 10% compared with LRU,and 39% compared with GreedyDualSize,and that TH-TS could reduce the amount of promoted data by 32% and 59%,and the amount of demoted data is also reduced by 47% and 66% compared with LRU and GreedyDualSize.