Date of Award

Spring 2010

Project Type


Program or Major

Computer Science

Degree Name

Doctor of Philosophy

First Advisor

Radim Bartos


In the first part of this dissertation, Data Over Cable Service Interface Specification (DOCSIS) and IEEE 802.3ah Ethernet Passive Optical Network (ETON), two access networking standards, are studied. We study the impact of two parameters of the DOCSIS protocol and derive the probability of message collision in the 802.3ah device discovery scheme. We survey existing bandwidth allocation schemes for EPONs, derive the average grant size in one such scheme, and study the performance of the shortest-job-first heuristic.

In the second part of this dissertation, we study networks of mobile sensors. We make progress towards an architecture for disconnected collections of mobile sensors. We propose a new design abstraction called tours which facilitates the combination of mobility and communication into a single design primitive and enables the system of sensors to reorganize into desirable topologies alter failures. We also initiate a study of computation in mobile sensor networks. We study the relationship between two distributed computational models of mobile sensor networks: population protocols and self-similar functions. We define the notion of a self-similar predicate and show when it is computable by a population protocol.

Transition graphs of population protocols lead its to the consideration of graph powers. We consider the direct product of graphs and its new variant which we call the lexicographic direct product (or the clique product). We show that invariants concerning transposable walks in direct graph powers and transposable independent sets in graph families generated by the lexicographic direct product are uncomputable.

The last part of this dissertation makes contributions to the area of storage systems. We propose a sequential access detect ion and prefetching scheme and a dynamic cache sizing scheme for large storage systems. We evaluate the cache sizing scheme theoretically and through simulations. We compute the expected hit ratio of our and competing schemes and bound the expected size of our dynamic cache sufficient to obtain an optimal hit ratio. We also develop a stand-alone simulator for studying our proposed scheme and integrate it with an empirically validated disk simulator.