Solving a Video-Server Load Re-Balancing Problem by Mixed Integer Programming and Hybrid Variable Neighborhood Search

Abstract

A Video-on-Demand system usually consists of a large number of independent video servers. In order to utilize network resources as efficiently as possible the overall network load should be balanced among the available servers. We consider a problem formulation based on an estimation of the expected number of requests per movie during the period of highest user interest. Apart from load balancing our formulation also deals with the minimization of reorganization costs associated with a newly obtained solution. We present two approaches to solve this problem: an exact formulation as a mixed-integer linear program (MIP) and a metaheuristic hybrid based on variable neighborhood search (VNS). Among others the VNS features two special large neighborhood structures searched using the MIP approach and by efficiently calculating cyclic exchanges, respectively. While the MIP approach alone is only able to obtain good solutions for instances involving few servers, the hybrid VNS performs well especially also on larger instances.