CDPFS: A Centralized Dynamic Parallel Flow Scheduling Algorithm to Improve Load Balancing on BCube Topology for Data Center Network

• Main Authors: Wei-KangChung, 鍾煒康
• Other Authors: Sun-Yuan Hsieh
• Format: Others
• Language: en_US
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/sqrkj7

碩士 === 國立成功大學 === 資訊工程學系 === 104 === In recent years, with the growing popularity of cloud computing, big-data applications increase demands for large and powerful data center networks (DCNs). There are more and more newly proposed network topologies for DCNs. BCube is a well-known recursively defined network structure which has good properties providing multiple low-diameter paths and good fault-tolerance for data center networks. Its distributed routing algorithm BSR which is fast and easy to build multiple parallel path set. But in the worst case, it may occur many flow collisions and waste 50\% capacity of each link in BCube. In this paper, in order to fully use those redundant links and improve load balancing in BCube, we add the central master computer to BCube topology and design a centralized dynamic parallel flow scheduling algorithm: CDPFS and CDPFSMP for single-path and multi-paths. We mainly focus on finding the least congestion path for each flow with global network status information and allocate those paths to each flow in parallel. The simulation result shows that our proposed algorithms which take advantage of BCube structure and give a good performance for load balancing.