Reducing Human Error in Network Configuration

Location: Yale University
Advisor: Y. Richard Yang

Configurations for today's IP networks are becoming increasingly complex. As a result, configuration management is becoming a major cost factor for network providers and configuration errors are becoming a major cause of network disruptions. Shadow Configurations is a novel technique providing realistic configuration evaluation (on the real network infrastructure) before deployment. We provide techniques for performance evaluation as well as a commitment protocol that offers network-wide transactional capabilities for network configurations. Shadow configurations is fully implemented in the Linux kernel, along with related performance testing and configuration commitment tools.

Improving Capacity and Efficiency in Wireless Mesh Networks

Location: Yale University
Advisor: Y. Richard Yang

Wireless Mesh Networks are inherently limited by interference between nodes caused by transmissions. A relatively recent development in this field is Cognitive Radio, which allows nodes to adapt their transmission schemes dynamically to use currently-unused parts of the radio spectrum. We would like to contribute to this research area as well as experiment with cognitive radio in a hardware implementation.

Analysis and its Practical Recommendations in Client Grids

Location: IBM
Collaborator: Nianjun Zhou

Client grid environments (e.g. SETI@home and World Community Grid) have been popular for computationally-intensive tasks. Despite their widespread use, a formal analysis of the mechanisms and design parameters used for redundancy and checkpointing is missing. This research begins this analysis with a basic model and to give some recommendations, then hopes to build it into a more accurate model and that encapsulates diverse client behavior. We would also like to analyze other factors that limit these environments, such as disk space at the management center and job completion times.

Clustering of Massive Financial Datasets

Location: Rensselaer Polytechnic Institute
Advisor: Petros Drineas

One of the major problems on Wall Street is designing new models for predicting stock performance. This research produced a software tool that imported historical stock prices from online data sources and acted as a testbed for various preprocessing methods that preceeded clustering.

Real-time Detection and Termination of Buffer Overflow Attacks

Location: Rensselaer Polytechnic Institute
Advisor: Christopher Carothers

Buffer overflow vulnerabilities in system software present a major security problem for operating systems. This research sought to intercept critical system calls via a loadable kernel module and verify their validity on-the-fly.