R&D: DLS, Delay-Life-Aware I/O Scheduler to Improve Load Balancing of SSD-Based RAID-5 Arrays

IEEE Xplore Digital Library has published, in 2019 IEEE 21st International Conference on High Performance Computing and Communications; IEEE 17th International Conference on Smart City; IEEE 5th International Conference on Data Science and Systems (HPCC/SmartCity/DSS) proceedings, an article written by Guohua Yan, Renhai Chen, Qiming Guan, and Zhiyong Feng, Tianjin University, China.

Abstract: “With the advent of the era of big data, efficient reading and writing of data have become the focus of attention. Solid State Drives (SSDs) have been a competitive alternative to traditional hard disks in disk arrays due of their better I/O performance. However, the uniform I/O scheduling strategies used in traditional disk arrays (e.g., FCFS) do not fully consider the characteristics of SSDs read and write performance asymmetry, resulting in insufficient utilization of disk array performance. Furthermore, a typical feature of a new type of solid-state drive based on new flash memory is the limited lifetime. How to balance the load while maximizing the lifetime of the flash SSDs have become the key to disk arrays research. To address the above issues, in this paper, we propose a novel delay-life-aware load scheduling strategy called ‘DLS’, which can optimize the I/O performance and prolong the lifetime of disk array. Firstly, due to different access times for different read/write request operations, we intelligently predict the delay time of each read and write request. Then, to improve the life of the disk array, we predict the current life of each SSD based on the load status of each SSD. Finally, based on the load delay and lifetime on each SSD, we design a write distribution model to dynamically predict the allocation address of the next write request. To demonstrate the effectiveness of delay-life-aware load scheduling strategy, we conduct experiments via a trace-driven simulator. The experimental results show that, compared with EGS, the IOPS and request execution time of the DLS scheduling strategy is improved by more than 8% compared with EGS. The proposed scheme can enhance the lifetime of the RAID-5 as its can allocate write requests more balanced than other scheduling strategies.“