Performance comparison and analysis of Linux block I/O schedulers on SSD

Abstract

A computer system’s one of the slowest operation is disk seek operation. Sending out read and write requests to the block devices such as disks as soon as the request arrives results in poor performance. After performing sorting and merging operations, the operating system kernel issues block I/O requests to a disk for improving the overall system performance. The kernel subsystem to perform scheduling the block I/O requests is named as the I/O scheduler. This paper introduces performance comparison and detailed analyses of Deadline, CFQ, Noop and BFQ block I/O schedulers that are contained in the Linux 4.1x kernel. The tests have been carried out on an SSD block device that is common in hardware combinations of both personal and professional use-case scenarios. The performance of the schedulers has been evaluated in terms of throughput. Each scheduler has advantages in different use-case scenarios and provides better throughput in a suitable environment.

Keywords

• Block I/O Scheduler,Deadline,Noop,CFQ,BFQ

References

1. R. Love, "The Block I/O Layer," in Linux Kernel Development, Crawfordsville, Indiana, Addison-Wesley, 2010, pp. 290-304.
2. B. Mao and S. Wu, “Exploiting request characteristics and internal parallelism to improve SSD performance,” in Proc. 33rd IEEE Int. Conf. Comput. Design (ICCD), NY, USA, Oct. 2015, pp. 447–450.
3. C. Ruemmler and J. Wilkes, “An introduction to disk drive modeling,” IEEE Comput., vol. 27, no. 3, pp. 17–28, Mar. 1994.
4. N. Agrawal, “Design tradeoffs for SSD performance,” in Proc. USENIX Annu. Tech. Conf. (USENIX ATC), Boston, MA, USA, Jun. 2008, pp. 57–70.
5. F. Chen, D. A. Koufaty, and X. Zhang, “Understanding intrinsic characteristics and system implications of flash memory based solid state drives,” in Proc. Joint Int. Conf. Meas. Model. Comput. Syst. (SIGMETRICS / Performance), Seattle, WA, USA, Jun. 2009, pp. 181–192.
6. B. Mao, S. Wu and L. Duan, "Improving the SSD Performance by Exploiting Request Characteristics and Internal Parallelism," in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 37, no. 2, pp. 472-484, Feb. 2018.
7. K. Shen and S. Park, “FlashFQ: A fair queueing I/O scheduler for flash-based SSDs,” in Proc. USENIX Annu. Tech. Conf. (USENIX ATC), San Jose, CA, USA, Jun. 2013, pp. 67–78.
8. C. Gao et al., "Exploiting Parallelism for Access Conflict Minimization in Flash-Based Solid State Drives," in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 37, no. 1, pp. 168-181, Jan. 2018.

9. W. Wang and T. Xie, “PCFTL: A plane-centric flash translation layer utilizing copy-back operations,” IEEE Trans. Parallel Distrib. Syst., vol. 26, no. 12, pp. 3420–3432, Dec. 2015.
10. F. Chen, R. Lee, and X. Zhang, “Essential roles of exploiting internal parallelism of flash memory based solid state drives in high-speed data processing,” in Proc. 17th Int. Conf. High-Perform. Comput. Archit. (HPCA), San Antonio, TX, USA, Feb. 2011, pp. 266–277.
11. Y. Hu, “Performance impact and interplay of SSD parallelism through advanced commands, allocation strategy and data granularity,” in Proc. 25th Int. Conf. Supercomput. (ICS), Tucson, AZ, USA, Jun. 2011, pp. 96–107.
12. H. Wang, “A novel I/O scheduler for SSD with improved performance and lifetime,” in Proc. 29th IEEE Symp. Massive Storage Syst. Technol. (MSST), Long Beach, CA, USA, May 2013, pp. 1–5.
13. W. D. Norcott, D. Capps, “Iozone filesystem benchmark,” [Online]. Available: www.iozone.org. [Accessed 4 October 2018].