Bandwidth adaptation for real-time streaming applications is typically designed to be conservative, since pushing for higher bandwidth could be counterproductive if it means an increased latency. However, such bandwidth adaptation operates based on the "symptoms" of congestion (e.g., increased delay) without knowing the underlying cause (self-congestion vs. cross-traffic). In this paper, we consider this problem in the context of Wi-Fi networks and introduce a novel technique, Ping-Pair, to measure and attribute congestion. We have integrated Ping-Pair into the popular Skype audio-video conferencing application to enable improved bandwidth adaptation dubbed Kwikr, using which we have conducted controlled experiments and also randomized A/B tests in a production setting.

Mobile communication is often frustratingly slow. When a user encounters poor performance, and perhaps even "confirms" the same by running a speed test, the tendency is to ascribe blame to the user's last-mile provider. However, as we argue in this paper, a more nuanced approach is needed to identify the location of the bottleneck responsible for the poor performance. Specifically, we focus on the question of whether the bottleneck lies in the cellular last hop (3G or LTE link) or elsewhere in the WAN path.

Real-time, interactive streaming for applications such as audio-video conferencing (e.g., Skype) and cloud-based gaming depends critically on the network providing low latency, jitter, and packet loss, much more so than on-demand streaming (e.g., YouTube) does. However, WiFi networks pose a challenge; our analysis of data from a large VoIP provider and from our own measurements shows that the WiFi access link is a significant cause of poor streaming experience.

The first wave of sensor network deployments from the early 2000s relied on aggregation-a strategy in which readings are combined locally using low-power radio links before they are communicated to the gateway. Aggregation reduced dependence on battery-draining, long-distance radio links, and reduced redundancy among reported data. We are now experiencing a second wave of sensor network research driven by ubiquitous smartphone usage. In this paper, we study the application of aggregation to the new smartphone sensor network setting, arguing that it can help reduce costs in contexts where existing cost-reduction strategies, such as opportunistic use of Wi-Fi and data piggybacking, do not apply. In more detail, we propose two new aggregation protocols, designed for the challenges of high mobility, that offer trade-offs in terms of bandwidth and energy savings. We then evaluate these protocols using both test bed experimentation (using a collection of 11 Samsung Galaxy Nexus smartphones running a Noise Tube-like application) and trace based simulation (using a large collection of mobility traces from taxi cabs in Singapore). Our experiments demonstrate that our aggregation protocols reduce cellular bandwidth usage by up to 95% while losing less than 5% of the data. Moreover, in many common cases, our protocols also yield significant energy savings.