BAHS: A Bandwidth-Aware Heterogeneous Scheduling Approach for SDN-Based Cluster Systems

摘要

Software Defined Networking(SDN) is a novel network architecture proposed by Stanford in 2008, which decouples control plane and data plane separately. SDN comes with logically centralized view of the whole network status and flexible control of packet switching, providing more possibilities of manipulating network in large-scale cloud systems. Scheduling algorithms impact the network performance significantly, especially for SDN-based cluster systems. However, finding an appropriate scheduling algorithm for heterogeneous network environments is an NP problem, where Quality-of-Service(QoS) needs to be ensured efficiently in these systems. Existing solutions rarely take into account of fine-grained bandwidth information and the lack of global network information also reasons why bandwidth is not widely used in scheduling. To address this problem, we proposed a new scheduling approach called BAHS, which is a comprehensive scheduling approach built in SDN environment. The main idea of BAHS is choosing proper tasks and their corresponding paths with the lowest interference. To estimate potential path conflicts, BAHS incorporates the state-of-the-art solution such as Dynamic Online Routing Algorithm(DORA). To demonstrate the effectiveness of our proposed approach, we conduct several simulations based on Mininet. Compared to previous approaches, BAHS improves the completion time and local rate of all tasks with QoS guarantee by up to 14.49% and 12.76%, respectively.