Luis Pedrosa’s research while affiliated with University of Southern California and other places

Large flows like videos consume significant bandwidth. Some ISPs actively manage these high volume flows with techniques like policing, which enforces a flow rate by dropping excess traffic. While the existence of policing is well known, our contribution is an Internet-wide study quantifying its prevalence and impact on video quality metrics. We developed a heuristic to identify policing from server-side traces and built a pipeline to deploy it at scale on traces from a large online content provider, collected from hundreds of servers worldwide. Using a dataset of 270 billion packets served to 28,400 client ASes, we find that, depending on region, up to 7% of lossy transfers are policed. Loss rates are on average six times higher when a trace is policed, and it impacts video playback quality. We show that alternatives to policing, like pacing and shaping, can achieve traffic management goals while avoiding the deleterious effects of policing.

As machine-to-machine networks become larger and more pervasive, manual configuration and discovery of resources will become intractable. It is in this context that we propose the RDL, a Resource Description Language that represents a uniform way of describing embedded resources, allowing them to be shared and enabling a new class of resource-aware applications. The RDL can describe a wide range of resources, characterizing individual nodes or entire networks. It can contribute to overcome performance issues in dense networks or mobility-driven problems in highly dynamic machine-to-machine topologies by providing the means for self-adaptability and manageability, as well as opportunistic resource sharing in context-aware embedded applications. The main goal for the RDL is to define a reusable and extensible resource description specification, which can only be reached if the resources are described in a standardized format. To illustrate the feasibility of our approach, we have also developed a Java implementation of the RDL framework, as well as a TinyOS implementation targeting resource constrained platforms. Furthermore, we have developed Modulus, a modular middleware for the development of resource-aware distributed applications.

Wireless sensing and actuation have been explored in many contexts, but the automotive setting has received relatively little attention. Automobiles have tens of onboard sensors and expose several hundred engine parameters which can be tuned (a form of actuation). The optimal tuning for a vehicle can depend upon terrain, traffic, and road conditions, but the ability to tune a vehicle has only been available to mechanics and enthusiasts. In this paper, we describe the design and implementation of CarMA (Car Mobile Assistant), a system that provides high-level abstractions for sensing automobile parameters and tuning them. Using these abstractions, developers can easily write smart-phone "apps" to achieve fuel efficiency, responsiveness, or safety goals. Users of CarMA can tune their vehicles at the granularity of individual trips, a capability we call personalized tuning. We demonstrate through a variety of applications written on top of CarMA that personalized tuning can result in over 10% gains in fuel efficiency. We achieve this through route-specific or driver-specific customizations. Furthermore, CarMA is capable of improving user satisfaction by increasing responsiveness when necessary, and promoting vehicular safety by appropriately limiting the range of performance available to novice or unsafe drivers.

From agriculture to industry, from the office to our homes, Wireless Sensor Networks (WSNs) are becoming a part of everyday life in many application areas. In typical WSN applications, the sensor nodes are fixed and interconnected amongst each other and to the outside world on a permanent basis. However, in certain types of applications, where the area to sense is wide and sensors are sparsely distributed, a different approach can be used. A mobile node can roam in the sensor field to collect and exchange information with disconnected clusters of nodes. This paper addresses the limitations of real world sensor networks with such moving nodes. To understand the behavior of a typical WSN node in these situations, two types of experiments were conducted. To begin with, the communication performance was measured, in a static scenario, establishing the base-line behavior. Afterwards, a second set of experiments was carried out with fast moving nodes at different speeds. Finally, the results of the two experiments were compared and analyzed.

A flexible Wireless Sensor Network platform has been developed and deployed at the Instituto Superior Tecnico/Technical University of Lisbon (IST-TUL). This test-bed integrates multiple projects into a single network, creating an expandable platform that facilitates future developments. To achieve this flexibility, a dedicated framework was developed, enabling fine-grained parameter control and application programmability through a centralised configuration panel. On top of this platform, four applications have been developed and currently coexist within the network, illustrating the new platform's capabilities. The article discusses the test-bed architecture and deployment challenges, as well as the environmental interaction and vibration monitoring applications. An experimental evaluation of these applications' capacity limits and performance shows that in-network processing must be used with the vibration monitoring application to avoid congestion collapse. Furthermore, a minimum amount of time is needed to complete processing tasks with the Environmental Interaction Application (approximately 200 ms in our test topology).

The main goal of this work is to characterise communications through body area networks (BANs), analysing how the human body affects wireless links qualities. 802.15.4 MICAz motes were deployed on seven on-body locations of four volunteers and a test-dummy. The quality of the communications was monitored in order to study the channel behaviour. Results show that the human body clearly modifies communication properties, introducing attenuations of up to 26 dB. Differences between male and female subjects were found to be unimportant. Body posture affects the performance, mainly because of body shadowing. LoS together with the channels dynamic nature are key issues to achieve connectivity. Power settings and the local environment also influence the BANs performance.

A new modular transport layer protocol for wireless sensor networks (WSNs) is proposed: WMTP - Wireless Modular Transport Protocol. This protocol not only allows the simultaneous use of all the main features commonly found in WSN transport protocols, namely congestion-control, fairness, and reliability, but also does so in a modular fashion. This way, the application layer can choose to use exactly the features that it requires. Additionally, WMTP also provides a set of uncommon features such as throttling, flow-control, transport layer quality- of-service, and optional integration with service-discovery. Focusing specifically on the fairness feature, experimental results show that the network resources are appropriately divided whether a simple fairness semantic is used, or in the presence of weighted differentiation. Additionally, this feature is shown to operate with minimal energy cost, as the protocol overhead is less than 10%.

A flexible wireless sensor network platform for easier implementation of diverse applications has been developed and deployed at the Institute Superior Tecnico - Technical University of Lisbon (IST-TUL). This test-bed integrates multiple projects into a single common network, thus creating an expandable platform that facilitates the development of future applications. To achieve this flexibility, a dedicated software framework was developed that not only provides a centralized configuration panel that is accessible over the Internet, allowing the administrator to configure common network parameters, but also supports application programmability, enabling fine-grained control of in-network sensing, processing, and actuation. On top of this platform, three initial applications have been developed and are currently coexisting within the same network, thus demonstrating the new platform's capabilities. The paper discusses the main issues related with the test-bed architecture and the development of an environmental interaction application, with an illustrative purpose, along with the deployment challenges. Results of the experimental evaluation of the test-bed are also shown, focusing on the performance of the environmental interaction application's in-network processing system. A particularly relevant result is denoted by the minimum time the network needs to complete its processing tasks (approximately 200 ms in our test topology).