2014 International Conference on Cloud Computing and Big Data (CCBD) (2014)
Nov. 12, 2014 to Nov. 14, 2014
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/CCBD.2014.17
The emergence of big data needs more and more storage capacity, and hard disk drive (HDD) plays a very important role in storage supplying. However, because of super paramagnetic effect, the growth of the areal density of HDD will quickly reach the limitation of 1Tb/in2. Shingled magnetic recording (SMR) is one of the most promising technologies to improve the areal density. In this paper, we have evaluated the performance of RAID system composed of SWDs. We have implemented a simulator, referred to as SWDsim, by extending Disksim-4.0 to simulate shingled write disk (SWD). We have proposed Shingle Translation Layer (STL) to SWDsim of which the main functions are address mapping and garbage collection. We describe our design and carry out extensive tests under enterprise, video monitoring and data archiving workloads on SWD-based RAID0/RAID10/RAID5. A set of lessons are extracted applicable to other SWD-based RAID systems. Our experimental results show that SWD-based RAID system has good spatial locality in data updates under read-dominant workloads or write-dominant workloads. And SWD-based RAID system shows nearly the same performance with HDD-based RAID system under read-dominant and sequential write-dominant workloads. However, mainly because of expensive garbage collection operations, SWD-based RAID system always performs worse than traditional HDD-based RAID system under write-dominant workloads with heavy data updates. Our early practice shows that SWDs are not suitable for RAID system deployment for enterprise applications which update data frequently, however, they could be used under read-dominant or sequential write-dominant applications such as cold data storage.
Time factors, Spatial databases, Monitoring, Arrays, Standards, Magnetic recording
W. Liu, D. Feng, L. Zeng and J. Chen, "Understanding the SWD-based RAID System," 2014 International Conference on Cloud Computing and Big Data (CCBD), Wuhan, China, 2014, pp. 175-181.