Exploring Mixed Integer Programming Reformulations for Virtual Machine Placement with Disk Anti-Colocation Constraints

TL;DR

This paper investigates different mixed integer programming reformulations for VM placement with disk anti-colocation constraints, aiming to reduce computation time and improve solution flexibility.

Contribution

It introduces two new reformulations of the VM placement problem and evaluates their effectiveness in reducing computation time compared to existing formulations.

Findings

Reformulations significantly affect computation time.

All three formulations are useful for different problem sizes.

Formulation COMB is particularly flexible and capable of solving large instances.

Abstract

One of the important problems for datacenter resource management is to place virtual machines (VMs) to physical machines (PMs) such that certain cost, profit or performance objective is optimized, subject to various constraints. In this paper, we consider an interesting and difficult VM placement problem with disk anti-colocation constraints: a VM's virtual disks should be spread out across the physical disks of its assigned PM. For solutions, we use the mixed integer programming (MIP) formulations and algorithms. However, a challenge is the potentially long computation time of the MIP algorithms. In this paper, we explore how reformulation of the problem can help to reduce the computation time. We develop two reformulations, by redefining the variables, for our VM placement problem and evaluate the computation time of all three formulations. We show that they have vastly different computation time. All three formulations can be useful, but for different problem instances. They all should be kept in the toolbox for tackling the problem. Out of the three, formulation COMB is especially flexible and versatile, and it can solve large problem instances.