Paper Title: Adapting STAR Code for Non-Volatile Memory Systems

Paper Abstract: Erasure codes have been widely used in various data storage products and systems, from disk arrays to data center systems. As non-volatile memory (NVM) technology emerges, I/O throughput of GBs/sec to 10s of GBs/sec is becoming reality. It thus calls for high performance erasure codes with much faster encoding and decoding speeds to match NVM disk's throughputs. In this paper, we evaluate and adapt computing performance of STAR Code, a class of erasure code that uses only XORs for encoding and decoding and can correct up to 3 erasures (disk failures), and compare its performance with traditional Reed-Solomon code currently being used in many storage systems. Extensive experimental measurements show that STAR with its parameters such as block size properly adapted for practical NVM disks and systems can outperform Reed Solomon code significantly in both encoding and decoding speed, e.g., on an i5-6500 Thinkcentre M900, STAR's can reach sustainable encoding speed at around 12GBs/sec, beating Jerasure by 4 to 5 times, and ISA-L by 50%; and one-erasure decoding speed at 25 to 30GBs/sec, 2 to 5 times faster than both Jerasure and ISA-L.

Paper Author: Rui Chen, PhD student, Wayne State University

Author Bio: Rui Chen is a PhD student at Wayne State University in Detroit, Michigan, USA. His advisor is Dr. Lihao Xu. His research interests include erasure code, fast memory computation and fast storage systems.