Improving Read Performance with BP-DAGs for Storage-Efficient File Backup

ABSTRACT

The continued growth of data and high-continuity of application have raised a critical and mounting demand on storage-efficient and high-performance data protection. New technologies, especially the D2D (Disk-to-Disk) deduplication storage are therefore getting wide attention both in academic and industry in the recent years. Existing deduplication systems mainly rely on duplicate locality inside the backup workload to achieve high throughput but suffer from read performance degrading under conditions of poor duplicate locality. This paper presents the design and performance evaluation of a D2D-based de-duplication file backup system, which employs caching techniques to improve write throughput while encoding files as graphs called BP-DAGs (Bi-pointer-based Directed Acyclic Graphs). BP-DAGs not only satisfy the ‘unique’ chunk storing policy of de-duplication, but also help improve file read performance in case of poor duplicate locality workloads. Evaluation results show that the system can achieve comparable read performance than non de-duplication backup systems such as Bacula under representative workloads, and the metadata storage overhead for BP-DAGs are reasonably low.

Keywords: