Stripped disk array organizations with distributed parity RAID-5: study and performance analysis

Abstract

This paper presents an examination and performance analysis of Redundant Array of Inexpensive Disks ver. 5 (RAID-5) and its comparison to the single disk drive performances. The focus is on the influence of the number of disks in RAID-5 to the globally achieved disk performances. Initially, we have applied the M/G/1 queue for the needs of analytically modeling a single disk performances. Then, we have set up an adequate analytical model for underlying file system. Finally, the proper workload analytical model has been defined and applied for four different RAID performance cases. Additionally, to provide a more comprehensive analysis, we have defined three types of workloads. The obtained test results have shown that there is a solid slowing in the RAID-5 performance, specifically for the workloads which are dominated by small updates. Increasing the number of disks has a different impact on the performances and depends on the workload. All benchmark results are interpreted based on a mathematical model of workload, file system access times and single disk access times.