Fast scalable optimization to configure service systems having cost and quality of service constraints
Jimmy Li, John W. Chinneck, Murray Woodside, Marin Litoiu · 2009
Large complex service centers must provide many services to many users with separate service contracts, while managing their overall costs. A scalable hybrid optimization procedure is described for a minimum-cost deployment of services on nodes, taking into account processing requirements and resource contention. This is a heuristic for a problem which is in general NP-hard. It iterates between a fast linear programming (LP) sub-problem, and a nonlinear performance model, both of which scale easily to thousands of services. The approach can be adapted to minimize cost subject to performance constraints, or to optimize a combined quality of service measure subject to cost constraints. It can be combined with tracked performance models to periodically re-optimize deployment for autonomic QOS management.