A Case Study of Performance Evaluation for RAID-Coded Storage Systems

摘要

RAID codes are extensive developed to supply high reliability and high available for storage systems in modern data center. For example, RAID-0, RAID-4, RAID-5 and RAID-6 are wide deployed in in-production storage systems, of which RAID-6 codes are popular with the ability for tolerate two-disk-failure. The paper focus on performance evaluation of storage systems powered by RAID-6, firstly propose an analytical model to measure encoding, decoding and updating complexity, which is translated into XOR operation numbers for storage systems, and quantify the complexity of RDP and RP-RDP (i.e., two RAID-6 data layout schemes) using the formulated analytical model. Computational complexity of RAID code can significantly influence the write performance of storage systems, i.e., optimal computational complexity achieve better storage performance. The numerical results show RP-RDP, with optimal updating complexity, achieve better I/O balancing than RDP. For example, RP-RDP speeds up the load balancing ratio of RDP by a factor of up to 2.95 in the case of Random-SW.