On endurance and performance of erasure codes in SSD-based storage systems

• A comprehensive analysis of performance and endurance of different erasure codes applied in SSD-based storage systems
• Reed-Solomon and EVENODD provide the best and the worst performance among the under study erasure codes, respectively.
• The best performance for erasure codes in SSD-based storage systems is achieved using smaller SUS.
• Impact of modifying SUS on the performance of different erasure codes is more significant than modifying NoD

Erasure codes are widely used in data storage systems to protect against disk failures. Employing erasure codes in an array of Solid-State Drives (SSDs) in storage systems necessitates designers to revisit different characteristics in comparison to Hard Disk Drives (HDDs), due to non-mechanical property of SSDs. One of the most important characteristics of SSDs is their limitation on the number of Program/Erase (P/E) cycles. By taking into account the characteristics of SSDs, this paper presents a comprehensive analysis to investigate the effects of three well-known erasure codes on the endurance and performance of SSD-based disk subsystems. The three erasure codes, i.e., Reed–Solomon, EVENODD, and RDP are implemented on the SSD-extension of DiskSim simulator. The results show that the endurance and performance of Reed–Solomon are on average 90% and 60% higher than other erasure codes, respectively. Additionally, the three erasure codes are compared in terms of different stripe unit sizes, number of disks, and various request sizes. The results show that configuring a disk array with a 4 KB stripe unit size will improve the endurance and performance of EVENODD by 1.8 × and 2.9 ×, respectively, as compared to 128 KB stripe unit size.