An SDN-Based Hyper File System

• Main Authors: Zheng-Ping Guo, 郭正屏
• Other Authors: Jehn-Ruey Jiang
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/649p7j

碩士 === 國立中央大學 === 資訊工程學系 === 104 === In this study, we propose a hyper file system, named Reef, which is a combination of the technologies of Software Defined Networking (SDN) and the Distributed File System (DFS). Reef uses the SDN controller to manage network traffic in a centralized way so that the network resources can be allocated more efficiently to reduce network latency, congestion, and bandwidth consumption. Furthermore, Reef uses DFS storage nodes to store data to keep high scalability; operators can increase or decrease storage resources dynamically according to users demands, leading to high utilization of the file system. Reef also integrates the erasure code (EC) to improve data reliability. The EC is a method to make data fault-tolerant; it divides data into different fragments, which are encoded with redundancy to be stored across a set of different locations. Compared with the traditional multi-replica method to make data fault-tolerant, the EC can utilize smaller data redundancy for higher data reliability. We also propose the Residual-Network Multiple Shortest Path (RNMSP) algorithm to find paths to the storage nodes minimizing the average end to end delay for storing EC-encoded data blocks. We simulate the RNMSP algorithm and other related storage node path-finding algorithms under different network topologies for comparing algorithms in terms of the execution time and the average end to end delay.