SDN for Dense Home Networks

Dense WiFi networks—where multiple APs and clients coexist—are commonplace and present certain challenges. In the US, 24% of households are in multiunit buildings, and world-wide a large and growing fraction of the population lives in high-density housing. Residential networks are chaotically managed and deployed. The vast majority of users install their WiFi APs with factory default settings, and lack the sophistication — or desire— to change the configuration. Multiple networks run independently and share the same airtime resources, which often leads to poor channel selection, poor coverage, and unsatisfying user experience. While better-managed and provisioned, enterprise networks face similar challenges. They deploy - often incrementally - a large number of APs, while trying to account for coverage, performance, mobility, and interference with neighboring networks.

Despite growing interest on improving dense networks both in enterprise and academia, we have little practical experience on how they really work. We are principally interested in studying and evaluating the pros and cons of new ways to control WiFi networks. For example, we plan to evaluate the benefits of centrally controlling a large WiFi network, controlling power, channel allocation and association from a single vantage point. To enable coordination of multiple APs when multiple owners/users are present, we introduce the notion of a "Personal Network". A "Personal Network" is an abstraction that decouples the underlying physical infrastructure from the logical view we expose to users. Using Personal Networks, we can extend enterprise-like WiFi network management to different contexts (e.g. a high-rise building, a residential neighborhood, or a city-wide WLAN infrastructure), while respecting the users' need to customize and personalize their own network service. We are working to refine and extend our Personal Network abstraction, as a generic way to expose a user-centric view of the network infrastructure.

We are exploring these ideas using a unique testbed we created with two main requirements: i) control how the wireless channel is used; and ii) real-world deployability in that our testbed should carry the traffic of real users, running any application they choose on any of their WiFi-connected devices (laptops, phones, tablets, DVRs, etc), and being able to apply their own policies for network control. We built and deployed a prototype network at a Stanford dorm, which acts as the primary network for ~40 students since January 2014. Early results include an up to 4x performance improvement comparing to the legacy network using different coordination schemes.

For more info, please contact: Yiannis Yiakoumis