Towards Efficient Cellular Traffic Offloading via Dynamic MPTCP Path Configuration with SDN

Cellular networks are suffering from the fast increasing volume of mobile traffic data. Hence, cellular network providers are eager to seek various ways to offload the cellular traffic to WLAN, which can enhance the quality of experience for their customers. Multi-path TCP provides the ability to utilize multiple communication channels simultaneously to transmit data packets so that it can be adopted to serve cellular traffic offloading to pursue higher average throughput than traditional single path communication. However, the greedy nature of MPTCP scheduler can lead to even worse performance compared to single path, which finally makes cellular traffic offloading less efficient sometimes. Therefore, motivated by the lack of practical solutions for operators to properly configure the communication path, we design and implement a system that can dynamically determine and deploy optimal path selections for mobile users leveraging on software-defined networking. Our system has two novel components: i) a support vector machine regression model to predict the potential performance of MPTCP and ii) an SDN controller continuously monitors the network status and dynamically adjusts path configurations for every user on demand. The goal is to maximize the network bandwidth resources for the cellular traffic offloading through WiFi with minimum changes on the network infrastructure. We build a Mininet-WiFi-based testbed to perform evaluation experiments. Results reveal that our system achieves seamless network handover and increases the average throughput by 8% without introducing extra overhead.