2009 IEEE International Conference on Networking, Architecture, and Storage (2009)
Zhang Jia Jie, Hunan, China
July 9, 2009 to July 11, 2009
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/NAS.2009.60
Performance and reliability are eternal topics in storage system. Nowadays, vast majority of storage system are made up of hard disk drive. The performance and reliability of hard disk drive are very important subclass of the former. So far, most researches of hard disk drive are single benefit. Some improve performance and some enhance the reliability. It is very little production which can heighten both performance and reliability at the same time. A new disk model we present fills this blank. The model not only improves disk drive performance, but also enhances reliability. The model is called MHPR (multi-head parallelism and redundancy). The models add more heads in one arm assembly, introduce redundancy and parallelism into disk. We analyze the performance and reliability in theory, and use different traces to evaluate the model in a simulator. We find that MHPR model can reduce seek time and transfer time effectively. The access time of single request is reduced. The model greatly shortens the queue time of single request in the case of two traces. So, the response of request is much faster; Response time of single disk drive is reduced by 98% and 80%, respectively in heavy workload and light workload. Using MHPR model to build RAID array, the MTTDL of this storage system can be improved 2-43 times.
storage; disk; multi-head; parallelism; redundancy
Y. Hu, D. Feng, J. N. Liu and S. P. Zhang, "MHPR: Multi-head Parallelism and Redundancy Disk Model," 2009 IEEE International Conference on Networking, Architecture, and Storage(NAS), Zhang Jia Jie, Hunan, China, 2009, pp. 315-322.