Energy Aware Flow Allocation Algorithm forFat-tree Data Center Networks

• Main Authors: Sung-Hsi Tsai, 蔡松熹
• Other Authors: Shang-Juh Kao
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/cgi-bin/gs32/gsweb.cgi/login?o=dnclcdr&s=id=%22107NCHU5394083%22.&searchmode=basic

碩士 === 國立中興大學 === 資訊科學與工程學系所 === 107 === Multi-rooted tree topology provides multiple parallel paths between host pairs. Therefore, it is very suitable as a network topology of a data center that needs to support cloud parallel computing. The routing method used by traditional data centers emphasizes the ability to make full use of bandwidth. For example, equal-cost multi-path routing(ECMP) splits traffic onto many equivalent paths for transmission to improve bandwidth usage. However, ECMP may cause unnecessary waste of electricity. Therefore, we propose an energy-aware path allocation mechanism under the software-defined network(SDN) architecture. SDN separates the control layer from the traditional network, and the controller centrally manages the path allocation of each flow. In our mechanism, the controller needs to detect the flow at the switches and establish a candidate path set for each new flow. After considering the overall network topology, assigning paths will minimize power consumption to new flow and turn off unused switches to reduce network power consumption. Each time new traffic is added, the port with the least power consumption is selected by the greedy energy-saving heuristic algorithm. However, the order in which each flow is added to the network affects the results of the heuristic algorithm. Therefore, we use the genetic algorithm to reroute the long-lived traffic in the network in a fixed cycle. At the end of the paper, we compare our method with the common routing strategy ECMP and greedy energy-saving heuristic. The experimental data shows that our method can reduce power consumption by 30%. Compared to greedy energy-saving algorithms that also have energy-awareness, our approach has about 7% of better power consumption reduction.