I am working with my advisor Dr. Karen Sollins on the Regions Project. A region is simply an abstraction that allows us to reason about arbitrary sets of nodes in the network that share common invariants, for instance peer-to-peer (P2P) members or nodes belonging to an autonomous system (AS). We are interested in providing a general architectural framework that allows reorganization and adaption within these regions. Based on discrete criteria (size, traffic patterns, latency, etc.) a region may reorganize itself continuously or after a threshold is passed using either centralized or distributed computation.
Below are initial results from using the idea of reorganization within both structured and unstructured P2P regions. Our research includes mechanism design and machine learning algorithms for reorganization. By explicitly including incentives within the system, our goal is to provide optimality and fairness in large distributed systems. We show in simulation promising results from our approach within this domain.
Disclaimer: These pages and documents are works in progress! Please do not distribute.