Zhang, Zhe2014-05-302014-05-302013-12https://hdl.handle.net/11299/163348University of Minnesota M.S. thesis. December 2013. Major: Electrical Engineering. Advisor: David Lilja. 1 computer file (PDF); vi, 38 pages.Erasure code, often applied in disk-array distributed le system, is a technique to protect data in case there are some accidental failures in the system. Nowadays, as customers require higher reliability for their data, RAID6, which at most protect double failures in disk-array, is not sucient any more. Triple failure protection scheme needs to proposed soon. RAID-DP(RDP) is one of most popular erasure codes on RAID6, and with close-form reconstruction, it can achieve optimal com- putational cost in terms of XORs. However, as the extension of RDP, close-form of RAID-TP(RTP) which provides triple failure protection can not achieve opti-mal computational cost. There is an alternatative en/decoding method based on matrix operations which seems promissing to replace close-form. Thus, we com- pared close-form en/decoding with matrix-based method, and we found that ma- trix en/decoding performs better only when the disk buer is small. Additionally, calculating decoding matrix cost too much time for reconstruction. Furthermore, we choose several dierent types of erasure codes in addition to RTP, and evaluate en/decoding performance with matrix method. The results show that RTP needs fewer XORs but it suers from calculating decoding matrix and updating small writes. Finally, we propose some potential techniques to address these issues that we may want to implement in future.en-USClose FormErasureMatrix reconstructionPerformanceThesis or Dissertation