Game-Aware and SDN-Assisted Bandwidth Allocation for Data Center Networks

摘要

Cloud computing has recently emerged as a promising paradigm for end-users and service providers. The application of the cloud-computing model to different applications offers many attractive advantages, such as scalability, ubiquity, reliability, and cost reduction to users and providers. By applying this model, the major computational parts of underlying applications are performed in data centers. Hence, effectively assigning the resources (e.g. memory, bandwidth) to applications plays a key role in providing a high Quality of Experience (QoE) to end-users. In the case of delay sensitive applications like video streaming and online gaming, the efficient resource allocation becomes more crucial. In this paper, we propose a game traffic friendly bandwidth utilization scheme using the Software Defined Networking (SDN) paradigm to solve the bandwidth allocation problem in cloud computing data center networks. Our proposed method makes use of machine learning techniques to classify the incoming traffic flows in real-time while ensuring game flows are prioritized over others. Our simulation results for a realistic network topology indicate good performance in terms of network traffic classification accuracy, and improvements of at least 9% in average utility (QoE), up to 30% increase in fairness (according to the Jain's fairness index), and on average an 8% reduction in delay experienced by users compared to a representative conventional method: Equal Cost Multi-path (ECMP).