Multi-Objective Virtual Machine Consolidation

Comparison of workload consolidation algorithms for cloud data centers

“…For example, Feller et al 23 used a weighted product, attaching the highest weight to the number of active PMs, while Jiang et al 10 considered an unweighted product of the total energy consumption and the SLA violation rate. Qiu et al 36 implemented a weighted function with the possibility to favor one or more metrics which shall be minimized (called Target) while ensuring that the other objectives do not become worse than before by using them as constraints (called Keep). The best solution is then chosen based on the ordered metrics inside Target from first to last.…”

“…Their objectives were to minimize VM migration costs, to maximize data center lifetime, to minimize the amount of service level agreement (SLA) violations and to minimize the energy consumption, so there is also the focus on avoiding unprofitable and aggressive reconfigurations. Qiu et al 36 considered CPU, RAM, and bandwidth as resources. Their objectives, for example, minimizing the number of used PMs or minimizing the number of migrations, are weighted and their algorithm used a ranking system for each individual.…”

“…There are also differences between the proposed approaches in the way they encode potential solutions as individuals in a population. The most used way to encode solutions of these algorithms is a VM‐to‐PM mapping, that is, specifying for each VM the PM that should host it 11,23,33,35,36 . A slightly different way is used by Hallawi et al: 9 each solution is represented by an array of structures, where each structure contains a host PM and an amount of VMs to be migrated from this PM.…”

“…As shown in Section 2, several population‐based metaheuristics have been suggested to solve the VMCP. We decided to implement three different and popular metaheuristics, namely, GA (for reference see 9, 36), ABC (for reference see 10), and PSO (for reference see 11). All these algorithms use a population (sometimes also called swarm) consisting of individuals where each individual encodes a possible solution of the VMCP.…”

A Vascular Invasive Tumor Growth Optimization Algorithm for Multi-Objective Optimization

“…The second type is to solve MOPs with prior knowledge (mostly appearing as weights), such as the multi-objective ant colony optimization [7], the variable length brain storm optimization algorithm [8], and the interactive preference-based multiobjective evolutionary algorithm [9]. The third type is to solve MOPs with some posteriori knowledge (mostly appearing as weights), such as the multi-objective virtual machine consolidation algorithm in paper [10]. The fourth type is to solve MOPs with interactive operations, such as the expert system set up online.…”