ncku-talk2013-02-21

NCTS(South)/ NCKU Math Colloquium

DATE 2013-02-21　16:00-17:00

PLACE R204, 2F, NCTS, NCKU

SPEAKER 洪暉智教授（交通大學工業工程與管理學系）

TITLE Pooling Effect: Separate or Not

ABSTRACT In this short talk, we attempt to give students in Mathematics Departments a brief introduction on the field of Operations Research (OR). The typical OR problem settings and the differences between OR and Mathematics are introduced first. We then focus on the complexity issues and discuss the solution approach method of heuristics and bounds.
For demonstration, we examine a stochastic resource allocation problem where servers are partitioned into multiple parallel work centers, and simultaneously job types are assigned to the work centers. Each job type has a distinct Poisson arrival rate and a distinct WIP weight. The goal is to minimize the total WIP cost. When servers are identical, each work center becomes a queueing system with an exponential service time distribution. A heuristic is constructed where the relative error is bounded above by 1 + e. When servers are non-identical, the server assignment discipline has a significant impact on the WIP cost. Heuristics are constructed for the random and fastest available server assignment disciplines. Theoretical bounds are developed for the relative errors of the heuristics. Also, heuristic performance is empirically evaluated.

NCTS(South)/ NCKU Math Colloquium

DATE	2013-02-21　16:00-17:00

PLACE	R204, 2F, NCTS, NCKU

SPEAKER	洪暉智教授（交通大學工業工程與管理學系）

TITLE	Pooling Effect: Separate or Not

ABSTRACT	In this short talk, we attempt to give students in Mathematics Departments a brief introduction on the field of Operations Research (OR). The typical OR problem settings and the differences between OR and Mathematics are introduced first. We then focus on the complexity issues and discuss the solution approach method of heuristics and bounds. For demonstration, we examine a stochastic resource allocation problem where servers are partitioned into multiple parallel work centers, and simultaneously job types are assigned to the work centers. Each job type has a distinct Poisson arrival rate and a distinct WIP weight. The goal is to minimize the total WIP cost. When servers are identical, each work center becomes a queueing system with an exponential service time distribution. A heuristic is constructed where the relative error is bounded above by 1 + e. When servers are non-identical, the server assignment discipline has a significant impact on the WIP cost. Heuristics are constructed for the random and fastest available server assignment disciplines. Theoretical bounds are developed for the relative errors of the heuristics. Also, heuristic performance is empirically evaluated.