A&S Departments and Programs   • Graduate Study Offered   American Studies • Anthropology • Applied Science • Art & Art History Biology • Black Studies Chemistry • Classical Studies Computer Science • Economics English Environmental Science & Policy Film Studies Geology Global Studies Government History • International Relations Kinesiology Linguistics Literary & Cultural Studies Mathematics Medieval & Renaissance Studies Military Science Modern Languages Music Neuroscience Philosophy Physics • Psychology • Public Policy • Religious Studies Sociology Theater, Speech & Dance Women's Studies

Search

Find People

Computer Science Department

Computer Science Department

New Theoretical Directions in Queueing Network Models for Capacity Planning

Giuliano Casale
College of William and Mary
Fri, Feb 2, 3:00 p.m., McGl 020

Product-form closed queueing networks are widely-used stochastic models for performance analysis and capacity planning of multi-tiered systems, storage servers, and distributed software systems. Despite the maturity of the field, several theoretical issues are still open, as the exact analysis of models with multiple service classes, which quickly becomes intractable as the job population grows.

In this talk, I will present the first efficient exact algorithm for multiclass networks with large populations. The algorithm is based on the solution of systems of linear equations involving normalizing constants of equilibrium state probabilitites, and promotes the application of linear algebra methods to queueing network analysis. I will show on simple examples that the algorithm can be several orders of magnitude faster than existing techniques and requires minimal storage.

The talk will then focus on a new class of inexpensive bounding techniques for performance metrics that can be applied in the adaptive optimization of computer systems. I will show that accounting for the role of non-bottleneck servers can double the accuracy of existing methods with similar computational cost.