I am working toward supporting a new generation of Internet services using structured overlay networks. This broad vision is outlined in an ICDCS 2017 vision track paper that describes how services that require highly demanding combinations of latency, reliability, resilience, and processing can be realized using the structured overlay concept.

My most recent work focuses on services with extremely low latency requirements (e.g. remote manipulation). This video shows me interacting with a Phantom Omni haptic device over a wide-area network. The signal from each device is sent halfway across the US before being sent back to the other device, so the latency is as if one device is on the East coast and the other is on the West coast. You can hear this latency when I tap on the desk. My work in this area aims to enable such low-latency communication and interaction with high reliability.

To support such applications, we have developed new overlay dissemination protocols that send messages over a subgraph of an overlay topology (a dissemination graph) to provide the necessary timeliness and reliability. This includes work with Emily Wagner, Michael Dinitz, and Yair Amir on constructing dissemination graphs that provide a good tradeoff between reliability and cost, which was selected for the best paper award at ICDCS 2017.

Before starting my PhD, I gained exposure to global-scale overlay technologies in the commercial world at LTN Global Communications. LTN is a cloud service provider that operates global overlay networks, transporting live video for the TV and media industries.

I am interested in building dependable infrastructure, or networked systems that maintain correct operation and predictable performance, even in the presence of partial failures or compromises. I am currently working on applying intrusion-tolerant principles to create SCADA systems for the power grid that can continue to operate correctly and at their required levels of performance even when part of the system has been compromised by a sophisticated attacker. In May 2017, we released version 1.0 of the Spire intrusion-tolerant SCADA system, which is designed to withstand malicious attacks at both the network level and the system level. Spire successfully withstood a red team attack conducted by Sandia National Laboratories at Pacific Northwest National Laboratory (PNNL) from March 27 to April 7, 2017 and was demonstrated in a test-deployment at the Hawaiian Electric Company from January 22 to February 2, 2018. Spire continues to be actively developed by the Resilient Systems and Societies Lab (RSSLab) at Pitt and the Distributed Systems and Networks (DSN) Lab at Johns Hopkins University.

I am additionally interested in consistent state maintenance in the presence of failures, including approaches such as Paxos and Extended Virtual Synchrony, as well as resilient (including intrusion-tolerant) systems more generally.

I developed a reliable, ordered multicast protocol based on a logical token ring that improves the state-of-the-art performance on 1-gigabit and 10-gigabit local area networks. The Accelerated Ring protocol circulates the token more quickly, reducing the impact of latency due to buffering and allowing for controlled parallelism in sending. I incorporated the Accelerated Ring protocol into the messaging protocol of the Spread Toolkit. A version of Spread that includes this protocol was released as an experimental version in July 2013, and the Accelerated Ring protocol is the toolkit's standard protocol for data center environments as of version 4.4.0.