Identifying the Input Region of Complex Queuing Models for a Preset Performance Threshold--An Efficient Simulation Methodology

• Activity Number: 80
• Type: Contributed
• Date/Time: Monday, August 9, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Ying-Chao Hung*+
• Companies: National Central University
• Keywords: network simulation ; input region ; performance threshold ; support vector regression ; space-filling design

Abstract

The development of network simulations has been largely devoted to improving the network performance over the last decade. In order to meet service level commitments, a procedure called Network Capacity Planning is often used to help adequately select input parameters for the simulation. The shortcoming of this type of simulation methodology is that, for large systems, it is extremely expensive both in terms of CPU time and use of available resources (e.g., processors). We present a methodological framework for designing efficient simulations for complex networks. The goal is to (1) adequately identify an input region such that the performance measure of interest (such as delay, backlog, jitter, etc) does not exceed a preset threshold, and (2) minimize the required number of simulation runs (experiment trials). Our approach works in sequential and combines the methods of SV (Support Vector) regression and space filling designs. A generalized switch model is used to illustrate the proposed methodology.