Virtual Machines Placement Problem Based on a Look-Ahead Workload Window over Distributed Cloud Gaming Infrastructure

EasyChair Preprint no. 1929

Abstract

Based on a distributed Cloud infrastructure, geographic region, available resources, delays constraints, power consumption, etc are factors that involve in determining the best datacenter where to allocate virtual resources. Allocation cost will also be determined according to that choice. For example, datacenters, installed in cold area, offer lower cost because they need few cooling maintenance. However, small datacenters, called Cloudlets, installed closer to players, could impose higher cost because of their limited resources or high need to cooling maintenance. The present paper contributes on optimizing resources allocation cost by proposing intelligent resources placement over a distributed Cloud infrastructure for Massively Multi-players Online Gaming (MMOG) service. We propose a dynamic Virtaul Machines (VMs) placement algorithm that enable inter migration of VMs between datacenters in order to guarantee low cost while respecting constraints of capacity and response delay. Our placement algorithm is based on a look-ahead window that predicts future workload of the Cloud service with the aim of minimizing number of VMs migration. Experiments show effectiveness of our contribution in maintaining the balance between low cost objective and delay and VMs migration constraints.

Keyphrases: distributed cloud infrastructure, MMOG, Multiple multi-dimensional knapsack problem, resources allocation, VMs placement problem

@Booklet{EasyChair:1929,
  author = {Eya Dhib and Khaled Boussetta and Nawel Zangar and Nabil Tabbane},
  title = {Virtual Machines Placement Problem Based on a Look-Ahead Workload Window over Distributed Cloud Gaming Infrastructure},
  howpublished = {EasyChair Preprint no. 1929},

  year = {EasyChair, 2019}}