Exploring patterns for scalability of network administration with topology constraints

Abstract

This thesis considers the impact of distributed network communication patterns on the scalability of dynamic systems configuration and monitoring using the software cfengine. Decentralized patterns are partially implemented as cfengine policy on topologies with node degree constraints. Experiments investigate total processing latency of patterns. Results show they provide a balanced approach to parallelization and scalability. The study of patterns on a chain topology reveals the challenge of phasing delay in deep tree structures. A time buffering method for reducing total processing latency is tested and found to be effective. Included are suggestions on new cfengine functionality and syntax to support patterns integration. As a whole, this thesis offers new perspectives in on-going patterns research as well as identifying challenges and solutions for bringing patterns to cfengine.