M.S. / Computer Science & Engineering / The Oregon Graduate Center's Cognitive Architecture Project (CAP) is developing a flexible architecture to evaluate and implement several types of neural networks. Wafer-scale integrated silicon is the targeted technology, allowing higher density and larger networks to be implemented more cheaply than with discrete components. The large size of networks implemented in wafer-scale technology makes it difficult to assess the effects of manufacturing faults on network behavior. Since neural networks degrade gracefully in the presence of faults, and since in larger networks faults tend to interact with each other, it is difficult to determine these effects analytically. This paper discusses a program, FltSim, that simulates wafer manufacturing faults. By building an abstract model of the CAP architecture, the effects of these manufacturing faults can be determined long before proceeding to implementation. In addition, the effects of architectural design trade-offs can be studied during the design process.

Thesis (Ph. D.)--University of Texas at Austin, 2003. / Supervisor: Mohamed G. Gouda. Vita. Includes bibliographical references. Available also from UMI Company.

Thesis (Ph. D.)--University of Texas at Austin, 2002. / Vita. Includes bibliographical references. Available also from UMI Company.

Thesis (M.S.)--University of Florida, 2002. / Title from title page of source document. Document formatted into pages; contains xi, 74 p.; also contains graphics. Includes vita. Includes bibliographical references.

Thesis (M.S. in Computer Science)--Naval Postgraduate School, June 2003. / Thesis advisor(s): Cynthia E. Irvine, Thuy D. Nguyen. Includes bibliographical references (p. 49-50). Also available online.

Delaunay triangulation (DT) is a useful geometric structure for net-working applications. We define a distributed DT and present a necessary and sufficient condition for a distributed DT to be correct. This condition is used as a guide for protocol design. We investigate the design of join, leave, failure, and maintenance protocols for a set of nodes in d-dimension (d > 1) to construct and maintain a distributed DT in a dynamic environment. The join, leave, and failure protocols in the suite are proved to be correct for a single join, leave, and failure, respectively. For a system under churn, it is impossible to maintain a correct distributed DT continually. We define an accuracy metric such that accuracy is 100% if and only if the distributed DT is correct. The suite also includes a maintenance protocol designed to recover from incorrect system states and to improve accuracy. In designing the protocols, we make use of two novel observations to substantially improve protocol efficiency. First, in the neighbor discovery process of a node, many replies to the node's queries contain redundant information. Second, the use of a new failure protocol that employs a proactive approach to recovery is better than the reactive approaches used in prior work. Experimental results show that our new suite of protocols maintains high accuracy for systems under churn and each system converges to 100% accuracy after churning stopped. They are much more efficient than protocols in prior work. To illustrate the usefulness of distributed DT for networking applications, we also present several application protocols including greedy routing, finding a closest existing node, clustering, broadcast, and geocast. Bose and Morin proved in 2004 that greedy routing always succeeds to find the destination node on a DT. We prove that greedy routing always finds a closest existing node to a given point, and our broadcast and geocast protocols always deliver a message to every target node. Our broadcast and geocast protocols are also efficient in the sense that very few target nodes receive duplicate messages, and non-target nodes receive no message. Performance characteristics of greedy routing, broadcast, and geocast are investigated using simulation experiments. We also investigate the impact of inaccurate coordinates on the performance of greedy routing, broadcast, and geocast. / text

published_or_final_version / abstract / toc / Electrical and Electronic Engineering / Master / Master of Philosophy

No description available.

To be effective, applications such as streaming multimedia require both a more stable and more reliable service than the default best effort service from the underlying computer network. To guarantee steady data transmission despite the unpredictability of the network, a single reserved path for each traffic flow is used. However, a single dedicated path suffers from single link failures. To allow for continuous service inexpensively, unreserved backup paths are used in this thesis. While there are no wasted resources using unreserved backup paths, recovery from a failure may not be perfect. Thus, a goal for this approach is to design algorithms that compute backup paths to mask the failure for all traffic, and failing that, to maximize the number of flows that can be unaffected by the failure. Although algorithms are carefully designed with the goal to provide perfect recovery, when using only unreserved backup paths, re-routing of all affected flows, at the same service quality as before the failure, may not be possible under some conditions, particularly when the network was already fully loaded prior to the failure. Alternate strategies that trade off service quality for continuous traffic flow to minimize the effects of the failure on traffic should be considered. In addition, the actual backup path calculation can be problematic because finding backup paths that can provide good service often requires a large amount of information regarding the traffic present in the network, so much that the overhead can be prohibitive. Thus, algorithms are developed with trade-offs between good performance and communication overhead. In this thesis, a family of algorithms is designed such that as a whole, inexpensive, scalable, and effective performance can be obtained after a failure. Simulations are done to study the trade-offs between performance and scalability and between soft and hard service guarantees. Simulation results show that some algorithms in this thesis yield competitive or better performance even at lower overhead. The more reliable service provided by unreserved backup paths allows for better performance by current applications inexpensively, and provides the groundwork to expand the computer network for future services and applications.

No description available.