The goal of this project is to build a distributed experimentation infrastructure spanning multiple sites in the US and Europe that is uniquely positioned to facilitate experimental research on global scale mobile services. The infrastructure would consist of cameras and touch-screen displays at different locations in New York and Madison as well as in CERTH and IBBT In Europe, allowing wireless based connectivity for its users through a layer 2 tunnel.
This project will enable research in distributed mobile services at a global scale. The research questions revolve around tradeoffs in placing computation across diverse locations around the globe and how they should be best replicated to optimize between performance latencies, bandwidth consumed and operational costs of such a service. The PIs are considering an application that they call a distributed wall where each site has a number or sensors and actuators, i.e. motion-detecting cameras and displays. A core local computational task is to quickly detect individuals and focus the cameras on them. As the number of sites grows, a key challenge for the application developers is to determine how and when to provision servers in different parts of the globe to optimize on the key experimental metrics. The proposed outcomes include better understanding of distributed deployment of global scale mobile services.
The project also has significant educational impact. The PIs will incorporate learnings from the project into the classroom through various networking and wireless communication courses at the respective universities. In addition, this project will bring students together from several countries to work together on a single project. The students will travel to the remote locations to give them exposure to research in different institutions across different countries with different research cultures and approaches to addressing networking research.
Both the U.S. and the E.U. have devoted considerable effort to the development of a research infrastructure ("testbeds") for experimenting with future Internet designs. However, it has not generally been possible before now to use resources - particularly resources belonging to wireless networking testbeds - from these parallel efforts in a single coordinated experiment. This functionality would allow researchers studying wireless networks to create innovative Internet designs on a truly global scale, to take advantage of the diversity of experimental resources, and to utilize resources in different time zones more efficiently. For this project, we have developed a software library which is used by experimenters to design and execute experiments across multiple U.S. and E.U. locations, including wireless testbeds at the Polytechnic Institute of New York University (US), the University of Wisconsin - Madison (US), Rutgers - the State University of New Jersey (US), the University of Thessaly (Greece), and iMinds (Belgium). Our library, "RemoteOmf," allows researchers to use a single RemoteOmf "experiment script" to coordinate and execute an experiment comprising elements across any combination of compatible testbeds on which they have registered, installed cryptographic keys for authorization, and reserved resources. We also developed an experimental use case for this library which examines the tradeoff involved in offloading computation-intensive tasks to "fog" or "cloud" computing elements. Wireless devices generally have limited energy and computing resources, so it is often desirable to offload certain computations to a more powerful server, either at the edge of the network close to the wireless device ("fog") or somewhere farther away ("cloud"). In the particular scenario we deployed, wireless devices determine their own location by sensing signal strength from all audible WiFi access points and/or WiMAX cellular towers and applying a Bayesian inference technique to determine their most probabilistic location. Considering latency and computing load throughout the system, the devices can elect to offload this computing task by sending the signal strength measurements for processing to either a nearby "fog" server at a testbed site on the same continent, or a faraway "cloud" server at a testbed site on another continent. This experiment showcases the benefit of the "RemoteOmf" framework, without which such an experiment would not be possible.
