My research tends to blend the areas of networking, security, games, and distributed systems. I am currently working on several projects, including the problem of large-scale consistency in peer-to-peer systems, cheat-proof protocols for multiplayer games, multiparty voice communications, real-time interaction in a virtual environment using motion capture (ie, augmented reality), and the modeling and simulation of virtual worlds.
As part of our work, we are developing secure and cheat-proof protocols for peer-to-peer games, whether it be for sharing state or sharing computations. Games are particularly challenging because players have an incentive to cheat while protocols must meet the real-time message passing requirements dictated by the type of game itself. We are currently using traffic traces and simulations to validate our protocols.
As part of our research in multiplayer games, we have studyied the characteristics of virtual populations in online games (see publications below). We have developed realistic simulation models for these types of games so that future architectures, whether they are client/server, peer-to-peer, or some hybrid, can accurately predict their viability. Our work now continues towards cheat-proof protocols for distributed computations in large-scale virtual spaces. We have also looked more recently at how to build large-scale trading card games in a peer-to-peer fashion. Source code and datasets for our measurements are freely available. Publications include:
We are investigating the characteristics of multi-party voice communication (MPVC) and their usage with games. In addition, one can easily see that MPVC is crucial to interaction in large-scale virtual spaces as it's the most natural method of communication between groups of people. At this point in our work, we have developed mathematical models for voice patterns and communication (see our published work below). We are now currently designing and developing secure protocols for distributed multiparty voice communication. Publications in this area include:
In one of our research classes, we developed a new system for multi-stream pipelines, which we named DUP. The goal of DUP is to provide a stream-based parallel programming paradigm that's simple and easy to use.
In an effort to show the efficacy of DUP in parallel programming, we have been working on a system for scale-free, multiparty video conferencing. Granted, we're not quite sure why you'd really want to scalably video-conference thousands of people simultaneously, but it's currently beyond the capability of modern systems to handle the computational load to do so. We use DUP and a set of filters that do parallel compression of the video streams on a GPU to get massive performance gains. We expect publications to soon follow describing our system.
Like many other researchers, we started investigating the issue of accurate collaborative filtering when the Netflix Prize provided a large data-set for experimentation. However, instead of aiming to win the prize, we have used the data-set to develop accurate methods for estimating Pearson's correlation coefficients.