WiFi offloading has emerged as a key component for improving network performance as one of the tools in the service provider's arsenal to address wireless data consumption. Traditionally, any WiFi has been viewed as a positive as WiFi offloads the user from the constrained cellular network onto what is likely be a better experience on WiFi network. The need for better Quality of Experience (QoE) leads many service providers to push users to join WiFi networks. Efforts such as ANDSF (Access Network and Discovery Service Function) and Hotspot 2.0 aim to streamline how to convey policy / awareness and simplify / hide complexity for WiFi association. Although there exist mechanisms in Hotspot 2.0 for the user device (User Element - UE) to query the WiFi access point (AP) with regards to performance, widespread deployment is still quite far into the future.
Although the research community has built various tools to pinpoint network performance (iperf) and capture mobile device performance, such tools tend to be quite heavyweight and are unable to deal with real-time decision making. The focus of our work is to deliver an order of magnitude improvement whereby we can correctly categorize WiFi performance on the order of hundreds of milliseconds (250 ms) with minimal bandwidth consumption (< 20 kB) and high classification accuracy. Our technique has applications for not only WiFi performance discovery for the purposes of association / non-association decisions but also for policy-level decisions from the cellular network (ex. downstream path selection from the trusted packet core) and client (ex. path loading for Multipath TCP).
The Fast Mobile Network Characterization aims to build a suite of tools to deliver such performance goals while doing so in a manner that does not require modification to the core of the client itself. We leverage inherent properties of TCP originally leveraged by Savage (TCP Sting) and expanded upon with our work on RIPPS (Rogue wireless Packet Payload Slicer) coupled with dynamic packet shaping (ex. work by Kurose). The net result is an approach that can be interfaced at a variety of levels ranging from in the web browser (Demo) to a dedicated program / app on the mobile device. One of the most intriguing service concepts under development is a technique whereby the visualization graph served on the webpage is the actual data flow that is served up in a carefully constructed manner. Similarly, we also are developing the equivalent of `performance GIFs' akin to tracking cookies but targeted at performance assessment (ex. invisible GIF).
Notably, it is our intention to open up the entirety of the FMNC platform whereby anyone can leverage and track their user performance on mobile devices. Hence, we will be making not only the code for FMNC available but all of the data that FMNC gathers as well through daily dumps of observed data.
If you are interested in collaborating on this project (researcher or industry), please drop a note to Prof. Aaron Striegel, the lead researcher for the project.
The demo server can be found here.