Efficient hybrid multicast approach in wireless data center network

• Main Authors: Zhu, Longting, Wu, Jigang, Jiang, Guiyuan, Chen, Long, Lam, Siew-Kei
• Other Authors: School of Computer Science and Engineering
• Format: Article
• Language: English
• Published: 2020
• Subjects: Engineering::Computer science and engineering; Wireless Data Center Networks; Multi-channel
• Online Access: https://hdl.handle.net/10356/141163
• Institution: Nanyang Technological University

Large-scale data center suffers from overload of data traffic on some bottleneck links, due to the fact that cloud-based services are mostly accomplished by group communications with multicast traffic. This paper investigates techniques of wireless transmission using multiple channels, instead of single available communication channel as reported in existing works, to enhance the flexibility of congestion control, and at the same time, considers load balance of different links. The objective is to meet the communication demands with as little as possible total data traffic, while simultaneously optimizes the load balancing among different links. The proposed two-stage framework jointly optimizes the transmission paths for both wireless and wired communications. The first stage models the interference relation via contradiction graph such that each node of the graph represents an alternative of a wireless link, and an edge indicates that two wireless links cannot coexist due to interference. As a result, the problem of finding the optimal arrangement for wireless communication without interference is transformed into the problem of finding the maximum independent set on the obtained contradiction graph. Also, an efficient strategy is proposed in this paper to construct contradiction graph from the wireless data center network, and to find the maximum independent set (MIS) whose elements indicates the destination nodes to be served by wireless links. The second stage develops a load balance aware routing algorithm to serve the remaining traffic demands that have not been served by wireless links. Experimental results show that, the proposed approach can significantly minimize total data traffic while simultaneously balancing the traffic loads of different links. The maximum load is reduced by up to 58.97%, 56.15% and 25.62% on Fat-Tree, VL2 and BCube based wireless DCNs, in comparison to the state-of-the-art.