원문정보
초록
영어
With the popularity of new storage technologies, hybrid active storage system provides an efficient way to improve the performance of high-performance computing applications. However, current active storage efforts have neglected the storage performance gap between heterogeneous servers, largely affecting the overall system performance. In this paper, we propose SDD, a Skewed Data Distribution scheme for hybrid active storage systems. In contrast to traditional even data distribution schemes, SDD distribute data on servers with skewed amount of data based on their performance. We have implemented a prototype of our proposed data layout scheme in a parallel I/O system, and demonstrated its benefits with a typical data processing application. Experimental results show our proposed data placement scheme can significantly improve the overall active storage system performance.
목차
1. Introduction
2. Related Work
2.1. Active Storage on Disk Devices
2.2. Active Storage on File Systems
2.3. Data Distribution in Parallel I/O Systems
3. The Skewed Data Distribution Scheme
3.1. The Basic Idea of SDD
3.2. Active Storage Data Processing Cost Model
3.3. Determination of Data Amount on Each Server
3.4. Skewed Data Distribution Scheme
3.5. Implementation
4. Performance Evaluation
4.1. Experimental Setup
4.2. Data Selection
5. Conclusion
Acknowledgements
References