Research

My main research interest focuses on large-scale distributed systems that span the entire Internet or consist of thousands of devices. The goal of my work is to support the architecture and engineering of scalable and robust Internet-wide applications by investigating new abstractions and infrastructures for building such systems. I am also interested in the unique algorithmic and data management challenges in this domain. In much of my previous work, I tried to build bridges between the areas of distributed systems, networking and databases.

Research Group

Current Research Projects

• Dependable Internet-Scale Stream Processing (DISSP): In the DISSP project, we investigate techniques for reliably processing large amounts of stream data coming from globally distributed sources. In order for such a global stream processing system to provide a robust service to millions of users, we develop approaches that degrade processing quality in a controlled fashion in response to resource shortages caused by failure or overload.
• SmartFlow: Current middleware is unable to adapt to the special requirements of healthcare applications in terms of auditing, controlled information flow, privacy and access control. The SmartFlow project investigates a lightweight architecture for building messaging middleware from a set of dynamic middleware extensions. Healthcare applications can express their requirements as extensions and push them into an intelligent middleware layer, simplifying applications design and improving performance.
• Pyxida: Network coordinates embed latency measurements between Internet nodes in a metric coordinate system. We have evaluated them as a means for estimating latencies between overlay nodes and performing nearest neighbour lookups on the Internet. We also propose new evaluation metrics and techniques for improving their accuracy and stability. In the Pyxida project, we run a public network coordinate service on PlanetLab that allows applications to use network coordinates for latency estimation. In addition, we experiment with geometric overlay routing using network coordinates.

Past Projects

• Stream-Based Overlay Network (SBON): An SBON is an overlay network that provides a common substrate for running multiple stream-processing applications on the Internet. It simplifies the construction of stream-processing applications by taking care of query instantiation, resource allocation, and dynamic query optimisation as network conditions change.
• Hermes: Hermes is a scalable event-based middleware that uses peer-to-peer techniques for the content-based routing of events in an overlay network of event brokers.
• DistCED: To ensure that event clients are not overwhelmed by the number of low-level events, DistCED is a service that allows the distributed detection of composite event patterns. DistCED uses extended finite state machines to detect event patterns on top of an existing publish/subscribe middleware.