PATTERN ENCODING ALGORITHMS AND INFORMATION MODELING METRICS FOR NETWORK QUALITY OF SERVICE

Author

Lucas John Vespa, Southern Illinois University CarbondaleFollow

Date of Award

5-1-2011

Degree Name

Doctor of Philosophy

Department

Electrical and Computer Engineering

First Advisor

Weng, Ning

Abstract

Networks are becoming increasingly complex, making network quality of service (QoS) an ongoing and difficult problem. Important QoS challenges include ensuring availability of services, privacy and accuracy of data and preventing data loss. These challenges can be addressed by providing effective security mechanisms that meet the scalability requirements of modern networks, environments for testing these mechanisms and providing metrics to measure and increase information quality especially in resource limited networks such as sensor networks. To meet these goals, we present several topics. First, to address the scalability problems of deep packet inspection (DPI) in network intrusion detection systems (NIDS), we theoretically characterize DFA in order to develop scalable pattern matching engines. Second, to keep up with the multi-gigabit rates of current and future networks and to increase DPI performance, we exploit the inter-stream parallelism of network traffic and the parallel processing capabilities of graphics processing units to create a multi-gigabit GPU based deep packet inspection engine. Third, we address the state explosion problem of multi-stride DFA and exploit intra-stream parallelism to achieve multiple Gb/s speeds of DPI on a single processor. Fourth, to evaluate large scale deployment of DPI we develop a software test-bed for evaluating worm containment systems. Lastly, we develop information quality metrics for sensor networks, and use these metrics to schedule sensor data collection and increase the quality of information.