Understanding performance bottlenecks of a multi-BSS software defined WiFi network testbed

WiFi networks have becoming popular in the past decade to provide network access due to its inexpensive but high-bandwidth infrastructure. With the increasingly dense WiFi deployment and the rapid proliferation of network devices and applications, the demand for more bandwidth, constrained delay performance and less power consumption is soaring; however, the traditional WiFi networks have become incapable to address various challenges such as unbalanced traffic load, handover and physical channel variation. In a WiFi network with multiple access points (AP), the traditional client association with APs is usually accomplished based on the distributed decisions such as the received signal strength, the earliest association reply etc. Lacking of coordination between APs, a WiFi network may suffer from unbalance traffic load. The emerging software defined WiFi networking (SDWN) introduces a centralized controller, which separates the control plane and the data plane, may also serve as a natural coordinator between APs. In this paper, we conducted a measurement study to understand the performance bottlenecks of a SDWN testbed. Our results demonstrate that SDWN has shown great potentials of flexible network management and control at the cost of performance degradation. We constructed this SDWN testbed based on the Odin project. The performance metrics examined in this study focus on the throughput and handover delay when mobile devices roam between APs. The virtual access points created in the SDWN effectively hide the handover delay for the mobile devices between APs; however, the throughput performance degrades by around 50% compared with the traditional WiFi networks. Our results show that there exists large design and optimization space for the current SDWN infrastructure.