An Attribute-Based Dynamic Data Organization in Mass Storage Systems

Chentao Wu, Changsheng Xie, Qinqi Wei, Shenggang Wan, Qiang Cao, Xubin He

Abstract: With the growing scale of storage systems, the single and static data organization strategies cannot satisfy the increasing demand of high performance and reliable I/O. One solution is to adjust data organization manually by administrators periodically, which causes high labor cost. This paper presents a dynamic data organization in mass storage systems based on an attribute-managed approach. The proposed attribute-managed approach describes and classifies the workload attributes in storage systems, and associating the workload attributes with certain data organization strategies. Storage objects are placed with the most appropriate data organization according to attributes pertaining to them. To demonstrate the effectiveness of our proposed solution, we developed a prototype of attribute-managed storage system by extending the existing OSD and iSCSI protocols. The measurement results show that the proposed attribute-based approach achieves high performance.