New Metrics for Characterizing and Predicting Network Behavior
Abstract
Networks are systems of point (nodes) with connections among some pairs of nodes measuring the degree of linkage. Networks represent an entire problem domain of many of the most difficult and unsolved mathematical problems. The objective of this effort was to formulate a foundational structure for networks and specifically develop new mathematical metrics for the description of networks in order to usefully describe both the static and dynamic properties of networks. Specifically, these new metrics provide a means of monitoring networks such as internet traffic over time by identifying anomalies, malicious processes, and abnormal network behavior. The criteria used for establishing network metrics were: (a) well-defined mathematically, (b) lossless in the description of a network, (c) hierarchical in providing a sequence of numerical metrics of decreasing importance, (d) intuitive in order to guide the use of the mathematical network expansions and associated metric values, (e) descriptive of the inherent topology of the network and strengths of connectivity, (f) sufficiently fast computationally in order to be dynamically useful as a tool, and (g) ideally distinguishing types of metrics that are: (1) network invariants, (2) variables which have dynamic behavior, and (3) variables which are chaotic or random. Results include: (a) finding network metrics that satisfy these criteria, (b) building the computer software to derive such metrics for general networks, and (c) testing limited internet traffic with this software.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2007
- Accession Number
- ADA462797
Entities
People
- Chin-tser Huang
- Cilia Farkas
- Duncan Buell
- Joseph E. Johnson
- Vladimir Gudkov
Organizations
- University of South Carolina