Performance evaluation of RAID6 systems
Yan Li, Tim Courtney, Roland N. Ibbett, Nigel P. Topham · File and Storage Technologies · 2007
The aim of this on-going work is to study the performance of RAID6 protected storage systems under a Storage Performance Council-1 (SPC-1) benchmark based workload [3]. With the scale of modern storage systems becoming increasingly large, the probability of having double disk failure in a protection group is much higher than before. As a result, there is an increasing demand for products protected against double disk failure. Although a number of double disk failure protection algorithms have been proposed, RAID6 has not been used extensively in practice due to its poor small write performance. In addition, there have been few publications about its performance under real world workloads. Given the trend of using RAID6 protection, it is important to understand the performance of RAID6 under real-world workloads. This work will study the performance of P+Q-like RAID6 algorithm as an exemplar (other RAID6 algorithms deploying two redundant disks in a protection group include Row-Diagonal Parity (RDP) [2] and EVENODD [1]) under an SPC-1 benchmark based workload using simulation. A discrete-event driven simulation model called SIMRAID has been developed to model the storage sub-system. SIMRAID contains a benchmark workload generator, a detailed RAID controller coupled with cache model, a transport-level Fibre-Channel model and disk model. The accuracy of SIMRAID has been verified through simulation of a system for which there are published SPC-1 benchmark results. This simulation shows a maximum inaccuracy of 5%. We use the maximum Business Scale Unit (BSU) value for which the average response time does not exceed 30ms as the performance metric.