Type of Document	Dissertation
Author	Thomas, Tavaris Jason
Author's Email Address	tavaristhomas@yahoo.com
URN	etd-11222004-120844
Title	Fire Ant: An Algorithm Providing High-Performance Distributed Fault-Tolerant Communication for Internet-Scale Network Survivability
Degree	Doctor of Philosophy
Department	Computer Science, Department of
Advisory Committee	Advisor Name Title Lois W. Hawkes Committee Chair Ernest McDuffie Committee Member Fred O. Simons Committee Member L. J. Kohout Committee Member
Keywords	Fault-Tolerance Swarm-Intelligence Data Communications Routing WAN
Date of Defense	2004-02-27
Availability	unrestricted
Abstract In today's world, data information, exchange, and processing are sustenance to corporations, organizations, and government agencies. Communications networks are relied upon to meet the computing needs of such organizations. These groups have computing demands that require large amounts of processing power, and use supercomputing systems consisting of massively parallel processors (MPP) for a wide range of data intensive operations. With advances in high-bandwidth network technology, the performance found in MPP can be replicated with a network of workstations (NOW). When large-scale disaster situations occur, intelligent routing algorithms are needed to maintain the necessary network resources required by NOW. These routing algorithms must be robust enough to accommodate changing network conditions and disaster situations while still providing resources for communications and acceptable load balancing of data flow. Mobile multiple autonomous units or multi-agents can be used to make routing and fault-tolerant decisions in a way similar to those of swarm insects. In this research, we develop a mobile multi-agent routing algorithm for disaster avoidance in network communications. This algorithm, called Fault-Tolerant Intelligent Routing Environment AntNet or FIRE Ant, is designed using activity structures. This is a novel approach for determining design parameters and requisite behavior of the routing algorithm. The activities of the NOWs environment are analyzed in order to find its relevant components. Activity structures are conceptual and methodological tools for capturing the adaptive activities in natural and artificial systems. In this research they allow the interaction of diverse structures in a high-performance, data communication network. The performance of the FIRE Ant algorithm is compared to a simplified version of the most well known mobile multi-agent based routing algorithm, AntNet, and to the BGP UPDATE procedure used in Internet routing. FIRE Ant is the first agent algorithm to address multiple network failures or disaster situations. FIRE Ant is a cost routing algorithm used to optimize the average flow of an entire network as a method of data flow load balancing during disaster situations. Network failures are added to the study in order to determine the effectiveness of the three algorithms during disaster situations. In the worst-case scenarios of this study, results show the FIRE Ant algorithm provides up to a 35.7% improvement in throughput over BGP UPDATE when a network is experiencing a disaster situation.
Files	Filename       Size       Approximate Download Time (Hours:Minutes:Seconds)     28.8 Modem   56K Modem   ISDN (64 Kb)   ISDN (128 Kb)   Higher-speed Access    FINAL-TAVARIS.pdf 3.30 Mb 00:15:16 00:07:51 00:06:52 00:03:26 00:00:17