ArticlePDF Available

Cloud computing resource allocation taxonomies

Authors:
Cloud Computing Resource Allocation Taxonomies
Fabio López-Pires
Itaipu Technological Park, National University of Asunción - Paraguay
fabio.lopez@pti.org.py
Benjamín Barán
National University of Asunción - Paraguay
bbaran@pol.una.py
Abstract Cloud computing datacenters dynamically provide millions of vir-
tual machines in actual cloud computing markets. Several challenging problems
have to be addressed towards an efficient resource management of these cloud
computing infrastructures. In the context of resource allocation, Virtual Machine
Placement (VMP) is one of the most studied problems with several possible
formulations and a large number of existing optimization criteria, considering so-
lutions with high economical and ecological impact. Based on systematic reviews
of the VMP literature, this work presents novel taxonomies in order to: (1) un-
derstand different possible environments where a VMP problem could be studied
from both provider and broker perspectives in different deployment architectures,
(2) identify existing approaches for the formulation and resolution of the VMP
as an optimization problem and (3) present a detailed view of the VMP problem,
identifying research opportunities to further advance in this area.
Keywords: Resource Allocation; Virtual Machine Placement; Cloud Computing;
Cloud Brokerage; Cloud Infrastructure; Optimization; Datacenters; Taxonomy.
1 Introduction
Cloud computing datacenters deliver infrastructure (IaaS), platform (PaaS) and software
(SaaS) as services, available to end users in a pay-as-you-go basis [1]. In this context,
a significant number of research challenges for delivering computational resources as an
utility have already been identified [2]. Achieving an efficient resource management in cloud
computing datacenters could be considered one of the most relevant challenges, including
important topics such as: resource allocation, resource provisioning, resource mapping and
resource adaptation [3]. Additionally, admission control and proactive elasticity could be
also considered relevant topics related to resource management in cloud datacenters [4].
The present work focuses on resource allocation, specifically in one of the most studied
problems for resource allocation in cloud computing datacenters: the process of selecting
which virtual machines (VMs) should be hosted at each physical machine (PM) of a cloud
Copyright © 2009 Inderscience Enterprises Ltd.
2Fabio López-Pires and Benjamín Barán
datacenter, commonly known as Virtual Machine Placement (VMP) problem. Several re-
search articles demonstrated that solving the VMP problem for efficient allocation of cloud
resources could significantly improve energy-efficiency, quality of service (QoS) and carbon
dioxide emissions; all of them with significant economical and ecological impact [5, 6, 7].
A conceptual cloud computing architecture considering the most studied problems in
cloud computing resource allocation is presented in [4]. According to the proposed archi-
tecture, the VMP is one of the main problems in the mentioned context, where placement of
admitted services could be solved considering different requirements and several criteria.
Beloglazov and Buyya proposed in [8] four different sub-problems for resource allo-
cation in cloud datacenters: (1) determining when a PM is overloaded, requiring migration
of VMs from this PM; (2) determining when a PM is underloaded, requiring migration of
all VMs from this PM and switching the PM to sleep mode; (3) selecting VMs to migrate
from an overloaded PM; and (4) finding a new placement of the VMs selected for migration
considering the overloaded and underloaded PMs.
The VMP problem has been extensively studied in cloud computing literature and several
surveys have already been presented. Existing surveys focused on specific issues such as:
(1) energy-efficient techniques applied to the problem [5, 9], (2) particular deployment
architectures where the VMP problem is applied, as federated clouds [10], and (3) methods
for comparing performance of placement algorithms in large on-demand clouds [11].
The above mentioned surveys and research articles focused into very specific issues re-
lated to the VMP problem. Consequently, López-Pires and Barán proposed in [12] a general
and extensive study of a large part of the VMP literature including more than 80 studied
articles [13], presenting a wide analysis of the existing approaches for the formulation and
resolution of the VMP as an optimization problem. Additionally, a novel taxonomy was
proposed in [12] for the classification of the studied articles by the main following criteria:
(1) optimization approach, (2) objective function and (3) solution technique.
The present work extends [12], complementing the proposed taxonomy with novel
taxonomies and presenting a detailed view of the existing approaches as well as several
possible research opportunities to further advance in this research area. The presented
taxonomies could guide interested readers to: (1) understand different possible environments
where a VMP problem could be studied, considering both provider and broker perspectives
in different deployment architectures, (2) identify existing approaches for the formulation
and resolution of the VMP as an optimization problem and (3) present a detailed view of the
VMP problem, identifying research opportunities to further advance in this research area.
The remainder of this paper is organized as follows: Section 2 presents a VMP pro-
blem environment taxonomy for the classification of related articles by: (1) orientation, (2)
deployment architecture and (3) types of formulation. Section 3 presents a VMP problem
formulation and resolution taxonomy considering the following criteria: (1) optimization
approach, (2) objective function and (3) solution technique. Section 4 details identified
research opportunities on this research area, while conclusions are left to Section 5.
2 VMP Problem Environment Taxonomy
Depending on the particular environment where a VMP problem will be studied, several
different considerations should be taken into account before considering a particular for-
mulation or technique for the resolution of the considered VMP problem. In this context,
Cloud Computing Resource Allocation Taxonomies 3
different possible environments could be identified by classifying research articles in the
VMP literature by: (1) orientation, (2) deployment architecture and (3) type of formulation.
First, a VMP problem could be studied from both provider or broker perspectives (see
Section 2.1). A provider-oriented VMP problem could be studied considering one of the fo-
llowing deployment architectures: single-cloud, distributed-cloud or federated-cloud, while
a broker-oriented VMP problem could be studied considering a multi-cloud deployment ar-
chitecture (see Section 2.2). Additionally, both provider-oriented and broker-oriented VMP
problems could be studied considering two different types of formulation: offline or online
formulations (see Section 2.3).
For a complete understanding of the possible environments where a VMP problem
could be studied, considering both provider and broker perspectives in different deployment
architectures, Figure 2.1 presents the taxonomy described in this section including relevant
references from the studied VMP literature [13].
The following subsections present a detailed description of the mentioned classification
criteria, including relevant definitions for a complete understanding of the VMP problem.
2.1 Orientations: Provider-oriented or Broker-oriented
Definition 2.1: A provider-oriented VMP problem is the process of selecting which VMs
should be hosted at each PM of a cloud computing infrastructure.
Resource allocation in cloud computing datacenters is a main concern for Cloud Service
Providers (CSPs). According to [12], the VMP problem is mainly formulated from this
perspective. It should be mentioned that in a provider-oriented VMP, a Cloud Service Tenant
(CST) cannot decide which PMs will host the requested VMs. In the specialized literature,
this particular problem is also known as Virtual Machine Allocation (VMA) problem [14].
Definition 2.2: A broker-oriented VMP problem is the process of selecting which VMs
should be hosted at each Cloud Service Provider (CSP) of a cloud computing market.
Considering that the number of CSPs has been rapidly increased and nowadays there
are different pricing schemes, VM offers and features, it is difficult for CSTs to search
for the most convenient option in cloud computing markets and decide which CSP is the
most convenient to host the requested resources. Consequently, a VMP problem can also
be formulated from a broker perspective who is responsible of helping CSTs to find good
allocation deals. In the specialized literature, this particular problem is also known as Cloud
Resource Brokerage (CRB) problem [15].
2.2 Deployment Architectures
Different deployment architectures can be considered for a VMP problem, depending on
the number of cloud computing datacenters associated to the problem instance and the type
of interconnection of these datacenters.
A provider-oriented VMP problem could be studied considering one of the following de-
ployment architectures: single-cloud, distributed-cloud or federated-cloud, while a broker-
oriented VMP problem could be studied in a multi-cloud deployment architecture.
4Fabio López-Pires and Benjamín Barán
VMP Problem
Provider-oriented
Single-Cloud
Offline
[16]
Online
[17]
Distributed-Cloud
Offline
[18]
Online
[19]
Federated-Cloud
Offline
RO
Online
[20]
Broker-oriented
Multi-Cloud
Offline
[21]
Online
RO
Figure 2.1:VMP Problem Environment Taxonomy. Relevant references for each envi-
ronment are presented. Unexplored environments are considered Research Opportunities
(RO).
Definition 2.3: A provider-oriented VMP problem in a single-cloud deployment is the
process of selecting which VMs should be hosted at each PMs of a single cloud computing
datacenter.
In scenarios considering a single cloud computing datacenter, a CSP could formulate a
VMP problem subject to commonly studied constraints such as: unique placement of VMs,
maximum capacity of resources [16] or Service Level Agreement (SLA) compliance [6, 22].
According to [12], the single-cloud deployment architecture is the most studied scenario in
the considered literature.
Definition 2.4: A provider-oriented VMP problem in a distributed-cloud deployment is
the process of selecting which VMs should be hosted at each PMs of more than one cloud
computing datacenter owned by the same Cloud Service Provider (CSP).
For CSPs with a global infrastructure (e.g. Amazon), a single-cloud deployment ar-
chitecture could be extended to several distributed cloud computing datacenters, where
the formulation of a VMP problem may include particular constraints and considerations.
CSPs with geo-distributed cloud computing datacenters may be interested in studying a
VMP problem for the provisioning of differentiated services to world-wide tenants, such
as redundancies of services (placement of VMs in different datacenters) [23] or quality
and performance of services (placement of VMs in datacenters closer to end users) [24],
increasing the complexity of the VMP problem formulation.
Definition 2.5: A provider-oriented VMP problem in a federated-cloud deployment is the
process of selecting which VMs should be hosted at each PMs of more than one cloud com-
puting datacenter owned by different Cloud Service Providers (CSPs) in a cloud federation.
In federated clouds, CSPs with excess capacity lease resources to other CSPs in need
of temporary additional resources; consequently, particular considerations associated to a
VMP problem in this deployment architecture have to be studied [10]. For example, trading
policies are generally not the same for CSPs that form part of the same cloud federation, so
Cloud Computing Resource Allocation Taxonomies 5
a VMP problem could be studied for cost-effective selection of CSPs for workload peaks,
just to cite a simple example.
Definition 2.6: A broker-orientedVMP problem in a multi-cloud deployment is the process
of selecting which VMs should be hosted at each Cloud Service Provider (CSP) of a cloud
computing market composed by more than one CSP.
Cloud Service Brokers (CSBs) or CSTs can require VMs to be deployed in cloud com-
puting datacenters owned by different CSPs, according to particular requirements such as
disaster recovery reasons or due to legislation [25], just to cite a few.
2.3 Types of Formulation: Offline or Online
A VMP problem could be formulated as an offline problem considering the placement of
VMs into PMs (for provider-oriented VMP) or VMs into CSPs (for broker-oriented VMP)
for a given needed requirement, i.e. a static service deployment.
Definition 2.7: An offline formulation of a provider-oriented VMP problem in any possible
deployment architecture could be understood as the process of selecting which VMs should
be hosted at each PMs of the considered cloud computing datacenters for a static cloud
service deployment.
For a provider-oriented VMP, an offline formulation does not consider possible re-
locations of VMs; therefore, there is no need for migration techniques to be applied. It should
be noted that offline formulations are mostly appropriate for initial placement of VMs or
for virtualized datacenters with deployments of VMs that rarely change its configuration
over time [26, 16, 27].
Definition 2.8: An offline formulation of a broker-oriented VMP problem could be under-
stood as the process of selecting which VMs should be hosted at each CSP of the considered
cloud computing market for a given (static) service deployment.
The broker-oriented VMP is mostly formulated as an offline problem, considering that
possible re-locations of VMs to different CSPs are still limited by interoperability factors.
A VMP problem could also be formulated as an online problem considering the place-
ment of VMs into PMs (provider-oriented VMP) or VMs into CSPs (broker-oriented VMP)
for a cloud service deployment with dynamic demand or parameters that change over time.
Definition 2.9: An online formulation of a provider-oriented VMP problem in any possible
deployment architecture could be understood as the process of selecting which VMs should
be hosted at each PMs of the considered cloud computing datacenters considering dynamic
demand or parameters that change over time.
It should be noted that an online formulation for a provider-oriented VMP problem could
be the most appropriate approach for resource allocation in cloud computing datacenters,
considering the on-demand model of cloud computing with dynamic resource provisioning
and dynamic workloads of modern cloud applications [8]. In this context, the authors of
this work are working on a detailed taxonomy for the classification of possible online for-
mulations for the provider-oriented VMP, presenting an analysis of possible challenges for
6Fabio López-Pires and Benjamín Barán
providers in cloud computing environments based on the most relevant dynamic parameters
studied so far in the existing VMP literature such as: (1) resource capacities of VMs (related
to vertical elasticity), (2) number of VMs of a service (related to horizontal elasticity) and
(3) utilization of resources of VMs (related to overbooking) [28].
Definition 2.10: An online formulation of a broker-oriented VMP problem in any possi-
ble deployment architecture could be understood as the process of selecting which VMs
should be hosted at each CSPs of a given cloud computing market considering demands or
parameters that change over time.
Considering that the number of CSPs has increased rapidly, nowadays several dynamic
parameters could be considered for the broker-oriented VMP problem such as: dynamic
pricing schemes, dynamic VM offers or dynamic requirements. Research on online formula-
tions of the broker-oriented VMP problem should advance to address existing opportunities
in emerging cloud computing markets as presented in a few articles [29, 30].
3 VMP Problem Formulation and Resolution Taxonomy
Considering each possible environment where a VMP problem can be studied (see Figure
2.1), several different formulations of the problem could be proposed. In this context,
formulations of a VMP problem may be classified by the: (1) optimization approach, (2)
objective function and (3) solution technique, as considered in [12].
First, a VMP problem could be formulated considering one of the following optimiza-
tion approaches: (1) mono-objective (MOP), (2) multi-objective solved as mono-objective
(MAM) or (3) pure multi-objective (PMO). Once the optimization approach is defined,
formulations may also be classified by the objective function(s) studied, both in minimiza-
tion and maximization contexts. These objective functions could be optimized separately
or simultaneously, depending on the selected optimization approach. Finally, solution tech-
niques for solving a VMP problem could be used as a third classification criterion [12].
For a complete understanding of possible approaches for the formulation and resolution
of a VMP problem, Table 1 presents the taxonomy described in this section, including exam-
ple references from the studied VMP literature [13]. The following subsections present a
detailed description of the classification criteria above defined as well as special conside-
rations for particular formulations.
3.1 Optimization Approaches
This section presents the optimization approaches considered in the articles studied in [12].
The identified optimization approaches may be classified as: (1) mono-objective (MOP),
(2) multi-objective solved as mono-objective (MAM) and (3) pure multi-objective (PMO).
The mentioned optimization approaches are detailed in the following subsections.
3.1.1 Mono-Objective (MOP)
A mono-objective optimization approach (MOP) considers the optimization of one objective
function or the individual optimization of more than one objective function, one at a time.
Research on the VMP problem has been mainly guided by this MOP approach, identify-
ing almost 40 different objective functions already proposed considering this optimization
Cloud Computing Resource Allocation Taxonomies 7
approach [12]. It should be noted that an objective function could be studied considering
different modeling approaches (e.g. economical revenue maximization could be achieved
by minimizing the total economical penalties for SLA violations [31], by minimizing ope-
rational costs [22, 32] or even by maximizing the total profit for leasing resources [33]).
3.1.2 Multi-Objective solved as Mono-Objective (MAM)
The optimization of multiple objective functions combined into one objective function is
considered in this work as a multi-objective problem solved as a mono-objective approach
(MAM). This hybrid approach enables the optimization of multiple objectives functions
with the disadvantage that it requires a deep knowledge of the problem domain to allow a
correct combination of the objective functions, which in most cases is not possible [34].
In the last years, a growing number of articles have proposed formulations of the VMP
problem with this hybrid optimization approach [13]. Different methods could be considered
for a formulation of a VMP problem with a MAM approach such as: weighted sum, solving
the problem as mono-objective while considering the other objective functions as constraints
of the problem or even proposing fuzzy logic to provide an efficient way for combining
conflicting objectives and expert knowledge [13].
3.1.3 Pure Multi-Objective (PMO)
A general Pure Multi-Objective Optimization problem (PMO) includes a set of pdecision
variables, qobjective functions, and rconstraints. Objective functions and constraints are
functions of decision variables. In a PMO formulation, xrepresents the decision vector,
while yrepresents the objective vector. The decision space is denoted by Xand the
objective space as Y. These can be expressed as [35]:
Optimize:
y=f(x) = [f1(x), f2(x), ..., fq(x)] (1)
subject to:
e(x)=[e1(x), e2(x), ..., er(x)] 0(2)
where:
x= [x1, x2, ..., xp]X(3)
y= [y1, y2, ..., yq]Y(4)
The set of constrains e(x)0defines the set of feasible solutions XfXand its
corresponding set of feasible objective vectors YfY. The feasible decision space Xfis
the set of all decision vectors xin the decision space Xthat satisfies the constraints e(x).
The feasible objective space Yfis the set of the objective vectors ythat represents the
image of Xfonto Y. These feasible spaces are defined as:
Xf={x|xXe(x)0}(5)
Yf={y|y=f(x)xXf}(6)
8Fabio López-Pires and Benjamín Barán
To compare two solutions in a pure multi-objective context, the concept of Pareto dom-
inance is used. Given two feasible solutions u,vX,udominates v, denoted as uv,
if f(u)is better or equal to f(v)in every objective function and strictly better in at least
one objective function. If neither udominates v, nor vdominates u,uand vare said to be
non-comparable (denoted as uv).
A decision vector xis non-dominated with respect to a set U, if there is no member
of Uthat dominates x. The set of non-dominated solutions of the whole set of feasible
solutions Xf, is known as optimal Pareto set P. The corresponding set of objective vectors
constitutes the optimal Pareto front P F [35].
Taking into account the large number of existing objective functions and possible approa-
ches for objective function modeling identified in [12, 13], PMO approaches could result
in more realistic formulations of a VMP problem, optimizing more than just one objective
function at a time (e.g. achieve economical revenue maximization by simultaneously mini-
mizing the total economical penalties for SLA violations, minimizing operational costs and
maximizing the profit for leasing resources). In this context, PMOs optimizing more than
three objective functions are known as Many-Objective Optimization Problems (MaOPs),
as defined in [57].
MaOPs differ significantly from PMOs because several issues should be considered
when solving optimization problems with more than three objective functions [58]. In case
of Pareto-based algorithms, these issues are intrinsically related to the fact that as the number
of objective functions increases, the proportion of non-dominated solutions grows, being
increasingly difficult to discriminate among solutions using only the Pareto dominance
relation [59]. Additionally, determining which solution to keep and which to discard in
order to converge toward the Pareto set is still a relevant issue to be addressed [58]. As the
number of objective functions grows, the proportion of non-dominated solutions to the total
number of solutions tends to one [60], making more difficult to solve a MaOP.
Table 1 VMP Formulation and Resolution Taxonomy. Example references for each environment
are presented. Unexplored environments are considered Research Opportunities (RO).
Technique Approach Objective Functions
f1(x)f2(x)f3(x)f4(x)f5(x)
Deterministic
Algorithms
MOP [36] [37] [31] [19] [38]
MAM [6, 39] [6, 40] RO RO [39, 40]
PMO RO RO RO RO RO
Heuristics
MOP [20] [41] [33] [42] [38]
MAM [43, 17] [43, 44] [44, 45] [17, 46] [47, 48]
PMO RO RO RO RO RO
Meta-Heuristics
MOP [49] RO [21] [50] RO
MAM [18, 51] [52, 18] [53, 51] [52] [53, 51]
PMO [16, 54] [16, 26] [16, 26] RO [54]
Approximation
Algorithms
MOP [55] RO RO RO RO
MAM [56] RO RO RO RO
PMO RO RO RO RO RO
Cloud Computing Resource Allocation Taxonomies 9
3.2 Objective Function Groups
In cloud computing datacenters or different markets, there are several criteria that can be
considered when selecting a possible solution for a VMP problem, depending on manage-
ment policies and optimization objectives. These criteria can even change from one period
of time to another, which implies a variety of possible formulations of the problem and
different objectives to be optimized.
According to [12], the VMP literature mainly focuses on the optimization of objective
functions that specifically concerns CSPs (provider-oriented VMP). Objective functions
could also be formulated considering the requirements of CSTs for allocation of a particular
service or application, often composed by more than one VM (broker-oriented VMP).
It should be mentioned that in [12], nearly 60 different objective functions were iden-
tified for the three optimization approaches presented in Section 3.1. Considering the large
number of proposed objective functions, identified objective functions with similar cha-
racteristics and goals were classified into 5 objective function groups that are presented
in the following subsections. Additionally, considering that MAM and PMO optimization
approaches may take into account any combination of each objective function group or
even different objective functions from the same objective function group, a simplified
classification is considered in this work (see Tables 1 and 2).
3.2.1 f1(x)- Energy Consumption
Energy consumption management is an important studied issue in the provider-oriented
VMP literature, with high impact in operational costs and carbon dioxide emissions for
cloud datacenter operations. According to [61], most of the time, servers operate in a very
low energy-efficiency possible region (i.e. between 10 and 50% of resource utilization), even
though energy efficiency is a very important issue to address, considering its economical and
ecological impact in modern datacenters. Energy consumption is the most studied objective
function for a VMP problem including different modeling approaches [12].
The most studied alternatives for modeling energy consumption include [12]: conso-
lidating VMs on the minimum number of PMs [62] and considering a linear relationship
between power consumption and Central Process Unit (CPU) utilization [5]. Joint optimiza-
tion of energy consumption and network traffic is also very studied in the VMP literature,
considering that network communication equipment represents between 10 and 20% of the
total datacenter energy consumption [56].
3.2.2 f2(x)- Network Traffic
As proposed in [5], network traffic is an important objective function for optimization with
open challenges in cloud computing datacenters for the provider-oriented VMP.
The main approaches for modeling network traffic include the optimization of: network
communication costs, live migration overhead, network metrics such as: delay, data access
and data transfer time, link congestion, network performance, service response time as well
as average latency, and Wide Area Network (WAN) communication in distributed-cloud
deployments [13].
A very studied approach for network traffic minimization is the placement of VMs
with high communication rate in the same PM (or at least in the same rack) to avoid
the utilization of network resources (or at least core network equipment). This approach
includes workload characterization and clustering techniques applied to a VMP problem
10 Fabio López-Pires and Benjamín Barán
to minimize the inter-VM network traffic [13]. Additionally, modeling and quantification
of live migration network overhead is an important open challenge in a provider-oriented
VMP network traffic optimization context [12].
Considering that VMs are dynamically created and destroyed in cloud computing en-
vironments, a consolidation process could require high level of flexibility where traditional
routing protocols present limitations to adjust flow paths. In [63], the authors proposed net-
work traffic load balancing to improve QoS in a VMP context considering Software Defined
Networks (SDN) [64], where flow paths are determined based on network status metrics
such as low delay, low packet loss or high security, just to cite a few.
3.2.3 f3(x)- Economical Costs
Economical costs optimization is an important studied issue in both provider-oriented and
broker-oriented VMP literature.
For a provider-oriented VMP, economical costs optimization is a key issue to be addre-
ssed and could be achieved by reducing operational costs. These operational costs are mainly
related to energy consumption minimization but other formulations could also be studied,
such as thermal dissipation costs [51]. Reducing penalty costs of SLA violations is another
studied approach in order to maximize the economical revenue of a CSP [13].
Finally, CSPs can maximize its economical revenue by leasing all its available resources
or at least the maximum possible [13]. In this context, VMP could be studied jointly with
admission control problem [4] and two possible scenarios could be identified: (1) if demand
for resources exceeds the current available resources, overbooking techniques or cloud
federation [4] can help CSPs to attend the requirements of the CSTs; on the other hand, (2)
idle resources can be offered in an auction-based scheme such as Amazon’s Spot Instances
[65]. Both scenarios represent open challenges for the VMP problem as well as emerging
cloud computing markets.
For a broker-oriented VMP, economical costs optimization is a key issue in actual cloud
markets and CSTs look for CSPs that meet the specific requirements of a particular service,
preferably with the minimal economical costs for the required cloud infrastructure.
The most studied pricing scheme in the considered VMP literature is fixed price. How-
ever, with the recent trend of dynamic pricing of cloud resources, where the price of resources
can vary depending on the free capacity and load of the CSP, few articles have recently
proposed formulations of a VMP problem considering dynamic prices schemes [66].
3.2.4 f4(x)- Performance
Performance optimization is an important studied issue in both provider-oriented and broker-
oriented VMP literature.
Performance modeling includes the optimization of: availability, CPU demand satis-
faction, deployment time, QoS, resource interference, security metrics, Shared Last Level
Cache (SLLC) contention and total job completion time [13]. Most of these performance
metrics may be considered for CSPs to improve the QoS in a provider-oriented VMP or for
CSTs in order to select an appropriate CSP to host their services in a broker-oriented VMP.
3.2.5 f5(x)- Resource Utilization
Cloud computing datacenters are commonly composed by different types of physical and
virtual resources such as CPU, RAM, storage, network bandwidth and even Graphical
Process Unit (GPU) in some cases.
Cloud Computing Resource Allocation Taxonomies 11
The main approaches include the maximization of resource utilization, but an important
issue to consider is the balanced utilization of each resource [13]. Li et al. studied the
concept of elasticity, referring to how well a datacenter can satisfy the growth of the input
VMs resource demands under limitations of PMs and network link capacities [67].
Considering the importance of the efficient utilization of resources, an interesting anal-
ysis of the anomalies and drawbacks in some existing strategies for efficient resource
utilization is presented in [68], proposing a novel vector-based technique that solves the
considered anomalies.
3.3 Solution Techniques
In the VMP literature considered in [12, 13], different techniques were considered for
solving a VMP problem. The main solution techniques include: (1) deterministic algorithms,
(2) heuristic algorithms, (3) meta-heuristic algorithms, and (4) approximation algorithms.
The four mentioned solution techniques are detailed in the following subsections.
3.3.1 Deterministic Algorithms
Classical deterministic techniques were considered for a VMP problem, including Con-
straint Programming (CP), Linear Programming (LP), Integer Linear Programming (ILP),
Mixed Integer Linear Programming (MILP), Pseudo-Boolean Optimization (PBO) and Dy-
namic Programming (DP) [12]. Most of these approaches are considered for solving novel
mathematical formulations of a VMP problem without any practical intention, considering
that obtaining the optimal solution implies a search in an universe of Npossible solutions
[16]:
N= (n+ 1)m(7)
where:
N: Size of the searching universe
n: Number of physical machines
m: Number of virtual machines
3.3.2 Heuristics
Considering that VMP is a combinatorial NP-complete problem [39], it is impracticable to
optimally solve instances of the problem for large number of PMs and VMs. Commonly,
CSPs are composed by thousands to millions of PMs and VMs, a scenario where optimal
solutions with exhaustive search algorithms can result extremely expensive. Therefore, a
trade-off between quality of solutions and computational cost has to be considered for real
world cloud management systems.
Heuristics have already been extensively studied in the literature for exponential com-
plexity problems. Several articles proposed heuristic-based solution techniques for a VMP
problem [12]. Most of the studied articles have proposed heuristics based on well-studied
algorithms such as: First-Fit, First-Fit Decreasing, Best-Fit, Best-Fit Decreasing, Worst-Fit
and Heaviest-Fit [13]. Other greedy algorithms were also proposed in addition to novel
heuristics proposed for the resolution of the VMP problem [13].
12 Fabio López-Pires and Benjamín Barán
3.3.3 Meta-Heuristics
As mentioned before, approximations to optimal solutions are sufficient in most of cloud
infrastructure environments. Meta-heuristics are also very useful in order to obtain good
solutions in practical time. Meta-heuristics include [12]: Memetic Algorithms (MA), Particle
Swarm Optimization (PSO), Ant Colony Optimization (ACO), Genetic Algorithm (GA),
Neighborhood Search (NS), Cut-and-Search,Simulated Annealing (SA) and Tabu Search
(TS) [13]. According to the proposed taxonomy, meta-heuristics are mainly studied with
multi-objective approaches (MAM and PMO) for solving VMP problems (see Table 1).
3.3.4 Approximation Algorithms
Heuristics and meta-heuristics provide good quality solutions, but the quality of the expected
solutions is hardly measurable. In a p-approximation algorithm, the value of a solution will
not be more (or less) than a factor ptimes the optimum solution. A small number of articles
proposed approximation algorithms for solving a VMP problem [12]. It is worth noting that
for cloud infrastructures, solutions obtained using heuristics or meta-heuristics techniques
are sufficiently good for most practical cases.
4 VMP Taxonomy: Research Opportunities
Based on an universe of 84 studied publications systematically chosen [13], Table 2 summa-
rizes the taxonomies presented in this work (see Figure 2.1 and Table 1). Considering that
5 studied articles [6, 38, 55, 56, 69] proposed 2 different solution techniques for the same
VMP formulation, statistics are based on 89 different VMP formulations (see Table 2).
Considering a detailed analysis of the information provided by Table 2, this section
summarizes research opportunities as well as on going research work by the authors in
order to further advance in this very active research area. It should be noted that statistics
presented in the following subsections are defined following the studied VMP literature (of
89 different VMP formulations) and considered only as a simple reference, given that more
articles may be considered and rapidly being published. Additionally, different systematic
reviews of the VMP literature (e.g. considering different keywords) or extended systematic
reviews (e.g. considering additional keywords) could result in different statistics.
Taking into account the diversity of existing environments, formulations and solution
techniques (see Table 2), it is important to provide a general vision on the main approaches
for the VMP problem, enabling interested readers to define specific research alternatives.
The following subsections describe research opportunities identified in this work as a main
result of the proposed VMP taxonomies (see Figure 2.1 and Tables 1 and 2).
4.1 Unexplored Environments, Formulations and Solution Techniques
Considering the studied VMP literature [13], unexplored VMP environments were identified
(see Research Opportunities (RO) in Figure 2.1). First, provider-oriented VMP problems
in federated-cloud deployments were not considered with offline formulations. Second,
broker-oriented VMP problems in multi-cloud deployments were not considered with online
formulations. No formulation or solution technique was studied for the mentioned unex-
plored VMP environments. Additionally, unexplored formulations and solutions techniques
were also identified in this work (see Research Opportunities (RO) in Table 1).
Cloud Computing Resource Allocation Taxonomies 13
Table 2 Virtual Machine Placement Taxonomy. Elements of column % represent the percentage
of articles in the studied universe [13]. For simplicity, MAM and PMO consider only the
number of considered objectives in column f(x). See Section 3.2 for f(x)details.
Oriented Deployment
Architecture Formulation Optimization
Approach f(x)Solution
Technique %
Provider
Single-cloud
Offline
MOP
f1(x)Heuristic 3.37%
Meta-Heuristic 2.25%
f2(x)Heuristic 2.25%
f4(x)Heuristic 1.12%
MAM 3Heuristic 2.25%
Meta-Heuristic 3.37%
2 Deterministic 1.12%
PMO 3 Meta-Heuristic 2.25%
2 Meta-Heuristic 1.12%
Online
MOP
f1(x)
Deterministic 4.49%
Heuristic 6.74%
Meta-Heuristic 1.12%
Approximation 1.12%
f2(x)Deterministic 2.25%
Heuristic 5.62%
f3(x)Deterministic 1.12%
Heuristic 5.62%
f4(x)Heuristic 7.87%
Meta-Heuristic 1.12%
f5(x)Deterministic 1.12%
Heuristic 4.49%
MAM
3Heuristic 3.37%
Meta-Heuristic 1.12%
2
Deterministic 3.37%
Heuristic 16.85%
Meta-Heuristic 1.12%
Approximation 1.12%
Distributed-cloud
Offline MOP f4(x)Deterministic 1.12%
Online
MOP
f2(x)Heuristic 1.12%
Deterministic 1.12%
f3(x)Heuristic 1.12%
f5(x)Heuristic 1.12%
MAM 3 Deterministic 1.12%
2 Meta-Heuristic 1.12%
Federated-cloud Online MOP f1(x)Heuristic 1.12%
Broker Multi-cloud Offline MOP f3(x)Deterministic 1.12%
Meta-Heuristic 1.12%
Total 100%
For MOP optimization approach, f2(x)(network traffic) and f5(x)(resource utilization)
were not studied considering meta-heuristics as solution technique. Additionally, approxi-
mation algorithms were studied as a solution technique only for f1(x)(energy consumption),
representing unexplored alternatives for the remaining studied objective function groups.
For MAM optimization approach, the resolution of VMP formulations considering
f3(x)(economical revenue) and f4(x)(performance) with deterministic algorithms were
14 Fabio López-Pires and Benjamín Barán
not studied. Similar to the MOP approach, approximation algorithms were studied as a so-
lution technique only for f1(x)(energy consumption), representing unexplored alternatives
for the remaining studied objective functions (see Research Opportunities (RO) in Table 1).
For PMO optimization approach, f4(x)(performance) was not studied considering
meta-heuristics as solution technique, while neither deterministic algorithms, heuristics nor
approximation algorithms studied PMO formulations of the VMP problem, representing
unexplored alternatives (see Research Opportunities (RO) in Table 1).
Taking into account a complete understanding of the VMP problem composed by VMP
environments and VMP formulations and solution techniques (see Table 2), several unex-
plored VMP problems could also be identified. In this context, Table 2 could guide interested
readers to identify existing research on VMP problems according to the studied VMP lit-
erature [13]. It should be noted that all possible VMP problems that are not presented in
Table 2 could be considered as a research opportunity. As an example, MAM and PMO
optimization approaches are not presented in Table 2 for the following VMP environments:
(1) provider-oriented VMP in distributed-clouds with offline formulations, (2) provider-
oriented VMP in federated-clouds with online formulations and (3) broker-oriented VMP
in multi-clouds with offline formulations.
4.2 Broker-oriented VMP considering Online Formulations
According to the proposed VMP Environment Taxonomy (see Figure 2.1), a broker-oriented
VMP problem could be studied in a multi-cloud deployment architecture considering offline
or online formulations. A broker-oriented VMP problem could also be studied conside-
ring a mono-objective approach (MOP) or multi-objective approaches (MAM or PMO),
optimizing different objective functions (see Table 1).
Considering that only 2.24% of the considered VMP literature studied a broker-oriented
VMP problem as an offline optimization problem (see Table 2), on going research by the
authors focus on exploring broker-oriented VMP formulations. In this context, an extended
systematic review of research articles demonstrate that broker-oriented VMP includes se-
veral research opportunities. The authors of this work are already working on novel math-
ematical formulations, specially for online broker-oriented VMP problems.
Several dynamic parameters could be studied for online broker-oriented VMP problems
such as: (1) pricing schemes, (2) VM offers or (3) user requirements, but a detailed survey on
these parameters and formulations is still needed (research opportunity). It should be noted
that the implementation of possible migrations of VMs across different CSPs is limited by
cloud interoperability factors and it is still out of the scope of this work.
4.3 Provider-oriented VMP considering Online Formulations
Considering the on-demand model of cloud computing, a provider-oriented VMP problem
should be solved dynamically to efficiently attend typical workload of modern applications.
According to the studied articles, 77.53% considered this particular type of VMP problem
with several different formulations and dynamic parameters.
On going research by the authors include detailed studies of the provider-oriented VMP
literature in order to understand possible challenges for CSPs in dynamic environments,
based on the most relevant dynamic parameters studied so far in the VMP literature. Pre-
liminary results identified that resource capacities of VMs (associated to vertical elasticity),
number of VMs of a cloud service (associated to horizontal elasticity) and utilization of
Cloud Computing Resource Allocation Taxonomies 15
resources of VMs (related to overbooking) are the most relevant dynamic parameters in
the specialized literature. Consequently, dynamic environments for online formulations of
the provider-oriented VMP problem could be classified by one or more of the following
classification criteria: (1) elasticity and (2) overbooking.
According to preliminary results obtained by the authors of this work, dynamic environ-
ments could be formulated considering one of the following elasticity values: no elasticity,
horizontal elasticity, vertical elasticity or both horizontal and vertical elasticity. Addition-
ally, identified dynamic environments may also consider one of the following overbooking
values: no overbooking, server resources overbooking, network resources overbooking or
both server and network overbooking.
Research opportunities for online formulations of a provider-oriented VMP include
detailed studies on complex dynamic environments (e.g. VMP considering both types of
elasticity and both types of overbooking) in order to enable CSPs to efficiently support
modern cloud applications and services. Considering that modern cloud services are often
composed by several inter-related VMs, the authors of this work are already working on
modeling techniques for these complex cloud services to consider any type of deployment
architecture (i.e. single-cloud, distributed-cloud, federated-cloud or multi-cloud).
4.4 Provider-oriented VMP considering PMO Optimization
According to the studied articles, only 3.37% considered a PMO approach for an offline
formulation of provider-oriented VMP problems simultaneously optimizing 2 or 3 objective
functions. PMO approaches could result in more realistic formulations of a VMP problem,
optimizing more than just one objective function at a time.
Taking into account that existing PMO formulations of a provider-oriented VMP pro-
blem consider at most 3 objective functions and more than 60 different objective functions
were identified in [12], several formulations could still be considered, specially for PMO
approaches. In this context, it is important to remember that PMOs optimizing more than
three objective functions are known as MaOPs. As described in Section 3.1.3, several issues
should be considered when solving optimization problems with more than three objective
functions [58]. Considering the mentioned challenges for solving MaOPs, López-Pires and
Barán have recently proposed in [27] a general many-objective optimization framework
that is able to consider as many objective functions as needed when solving an offline VMP
problem in a PMO context. In order to converge to a treatable number of non-dominated
solutions, the authors proposed the utilization of interactive lower and upper bounds asso-
ciated to each objective function to reduce the number of possible solutions of the Pareto
set approximation Pknown , following a decision maker needs.
Additionally, online formulations of a provider-oriented VMP considering a PMO
approach should be explored. For this particular type of VMP problem, PMO approaches
should include strategies for an appropriate solution selection from the Pareto set approxi-
mation Pknown , composed by non-dominated solutions. In this context, Ihara et al. already
proposed a first formulation of a Many-Objective VMP problem (MaVMP) for dynamic
environments, presenting studies on the evaluation of several automatic selection strategies
for provider-oriented VMP problems formulated as MaOPs [70].
16 Fabio López-Pires and Benjamín Barán
4.5 Provider-oriented VMP in Distributed and Federated Clouds
According to the studied articles, only 7.87% proposed the provider-oriented VMP problem
in a distributed-cloud deployment architecture, while only [20] studied the provider-oriented
VMP problem considering a federated-cloud deployment architecture (see Table 2).
Resource allocation in distributed and federated cloud environments is an active research
area [71]. In a provider-oriented VMP context, several particular constraints and objective
functions may still be studied and evaluated in order to fulfill the requirements of a CSP
with geo-distributed cloud computing datacenters or different CSPs in a cloud federation.
It should be noted that a broker-oriented VMP in a distributed-cloud deploy-
ment architecture may be adapted as a multi-cloud deployment architecture by con-
sidering each datacenter as a different CSP in order to fulfill requirements of high-
availability or legal issues (e.g. Amazon’s us-east-1, us-west-2 or eu-west-1 datacenters)
[https://aws.amazon.com/about-aws/global-infrastructure].
For a VMP problem in a distributed or federated-cloud deployment architecture, several
research opportunities may be still proposed considering unexplored objective functions,
novel formulations in MOP, MAM or PMO approaches or even experimental evaluation of
different solution techniques.
5 Conclusions and Future Directions
Based on an universe of 84 studied articles with 89 different VMP formulations [13], this
work presented general taxonomies of the VMP problem (see Table 2) considering possible
environments where the VMP problem could be studied (see Research Opportunities (RO)
in Figure 2.1) as well as different possible formulations and techniques for the resolution
of the VMP problem (see Table 1).
Depending on the particular environment where a VMP problem will be studied, se-
veral different considerations should be taken into account before considering a particular
formulation or technique for a VMP problem resolution. In this context, different possible
environments were identified by classifying research articles in the VMP literature by: (1)
orientation, (2) deployment architecture and (3) type of formulation.
A VMP problem could be studied from both provider or broker perspectives (see Section
2.1). A provider-oriented VMP problem could be studied considering one of the follo-
wing deployment architectures: single-cloud, distributed-cloud or federated-cloud, while
a broker-oriented VMP problem could be studied considering a multi-cloud deployment
architecture (see Section 2.2). Additionally, both a provider-oriented and a broker-oriented
VMP problem in any of the possible deployment architectures could be studied considering
two different types of formulation: offline or online formulations (see Section 2.3).
Considering each environment where a VMP problem was studied (see Figure 2.1),
several possible formulations of the problem could be proposed. In this context, formulations
of a VMP problem was classified by: (1) optimization approach, (2) objective function
and (3) solution technique. First, a VMP problem could be formulated considering one
of the following optimization approaches: (1) mono-objective (MOP), (2) multi-objective
solved as mono-objective (MAM) or (3) pure multi-objective (PMO). Once the optimization
approach is defined, formulations were classified by the objective function(s) studied, both
in minimization and maximization contexts. These objective functions could be optimized
Cloud Computing Resource Allocation Taxonomies 17
separately or simultaneously, depending on the selected optimization approach. Finally,
solution techniques for solving a VMP problem were used as a classification criterion [12].
According to Table 2, the VMP problem have been mostly studied as an online pro-
blem from the provider perspective, considering a single-cloud deployment architecture,
representing 69.66% of the studied articles. For this particular VMP environment, 42,70%
considered a MOP approach. It could be said that online formulations of provider-oriented
VMP problems in single-cloud deployment architecture is a well studied problem and ex-
isting literature should guide CSPs to solve VMP problems with these considerations.
Based on the proposed taxonomies, several research opportunities were identified in
the following research directions (see Section 4): (1) unexplored environments, formula-
tions and solution techniques, (2) broker-oriented VMP considering online formulations,
(3) provider-oriented VMP considering online formulations, (4) provider-oriented VMP
considering PMO optimization and (5) provider-oriented VMP in distributed and feder-
ated clouds. It should be mentioned that other relevant research opportunities could also be
identified with the proposed taxonomies.
Focusing on the large number of identified objective functions [12], the following ques-
tions still have no answer considering the studied VMP literature [13]: (1) can CSPs effi-
ciently optimize more than 3 objective functions for the VMP problem in a reasonable time?,
(2) which solution technique is the most appropriate for solving VMP problems considering
PMO approaches considering real scenarios?, (3) are PMO approaches the best choice for
online formulations or MAM approaches have the potential of performing even better?.
Additionally, the following questions focus on the variety of possible deployment archi-
tectures where the VMP problem could be studied: (1) how can CSPs model and formulate
and solve VMP problem in distributed or federated clouds?, (2) which solution technique
is the most appropriate for broker-oriented VMP for large scale customers?, (3) which
objective functions should a CSP consider in different deployment architectures?.
In order to answer the above mentioned relevant questions, research should focus on
unexplored formulations of the VMP problem, developing novel techniques and providing
methods and accepted benchmarks to compare and evaluate different approaches. Address-
ing this unexplored formulations may start applying an extended review of the VMP litera-
ture, considering that this work studied a relevant sub-set of the existing VMP literature in
order to guide interested readers, providing a general vision on this research area.
Additionally, future directions may include detailed studies on the VMP problem con-
sidering dynamic pricing schemes or auction-based pricing. These type of pricing schemes
represent open challenges for both CSPs and CSBs in the VMP research and emerging cloud
computing markets. Techniques for avoiding an excessive number of possible placement
combinations is also an open challenge for the VMP problem [52].
It is important to mention that actual cloud markets are mostly composed by thousands
to millions of VMs which are dynamically created and destroyed, so experimental tests for
VMP problem should consider: (1) large number of VMs and PMs, (2) heterogeneity in
PMs and VMs configurations, (3) diverse types and workload distribution, and (4) trending
dynamic parameters. Considering the studied articles, there is no test problems for the VMP
that could be used as a world accepted benchmark. A test problem generator could be very
useful at the time of composing experimental alternatives, giving the difficulties to access
real traces of world class cloud computing companies. As a general conclusion, it could be
said that different methods and algorithms should be still evaluated before a real good tool
is ready for massive use in commercial cloud computing datacenters.
18 Fabio López-Pires and Benjamín Barán
References
[1] Peter Mell and Tim Grance. The NIST definition of cloud computing. National Institute of
Standards and Technology, 53(6):50, 2009.
[2] Rajkumar Buyya, Chee Shin Yeo, Srikumar Venugopal, James Broberg, and Ivona Brandic.
Cloud computing and emerging it platforms: Vision, hype, and reality for delivering computing
as the 5th utility. Future Generation computer systems, 25(6):599–616, 2009.
[3] Sunilkumar S Manvi and Gopal Krishna Shyam. Resource management for infrastructure as a
service (iaas) in cloud computing: A survey. Journal of Network and Computer Applications,
41:424–440, 2014.
[4] Erik Elmroth, Johan Tordsson, Francisco Hernández, Ahmed Ali-Eldin, Petter Svärd, Mina
Sedaghat, and Wubin Li. Self-management challenges for multi-cloud architectures. In Towards
a Service-Based Internet, pages 38–49. Springer, 2011.
[5] Anton Beloglazov, Jemal Abawajy, and Rajkumar Buyya. Energy-aware resource allocation
heuristics for efficient management of data centers for cloud computing. Future Generation
Computer Systems, 28(5):755–768, 2012.
[6] Ankit Anand, J Lakshmi, and SK Nandy. Virtual machine placement optimization supporting
performance SLAs. In Cloud Computing Technology and Science (CloudCom), 2013 IEEE 5th
International Conference on, volume 1, pages 298–305. IEEE, 2013.
[7] Rajkumar Buyya, Chee Shin Yeo, and Srikumar Venugopal. Market-oriented cloud computing:
Vision, hype, and reality for delivering it services as computing utilities. In High Performance
Computing and Communications, 2008. HPCC’08. 10th IEEE International Conference on,
pages 5–13. Ieee, 2008.
[8] Anton Beloglazov and Rajkumar Buyya. Optimal online deterministic algorithms and adaptive
heuristics for energy and performance efficient dynamic consolidation of virtual machines in
cloud data centers. Concurrency and Computation: Practice and Experience, 24(13):1397–1420,
2012.
[9] Leili Salimian and Faramarz Safi. Survey of energy efficient data centers in cloud comput-
ing. In Proceedings of the 2013 IEEE/ACM 6th International Conference on Utility and Cloud
Computing, pages 369–374. IEEE Computer Society, 2013.
[10] Monica Gahlawat and Priyanka Sharma. Survey of virtual machine placement in federated
clouds. In Advance Computing Conference (IACC), 2014 IEEE International, pages 735–738.
IEEE, 2014.
[11] Kevin Mills, J Filliben, and Christopher Dabrowski. Comparing vm-placement algorithms for
on-demand clouds. In Cloud Computing Technology and Science (CloudCom), 2011 IEEE Third
International Conference on, pages 91–98. IEEE, 2011.
[12] Fabio López-Pires and Benjamín Barán. A virtual machine placement taxonomy. In Proceedings
of the 2015 IEEE/ACM 15th International Symposium on Cluster, Cloud and Grid Computing.
IEEE Computer Society, 2015.
[13] Fabio López-Pires and Benjamín Barán. Virtual machine placement literature review. Technical
report, Polytechnic School, National University of Asunción, 2015.
[14] Ryan Jansen and Paul R Brenner. Energy efficient virtual machine allocation in the cloud. In
Green Computing Conference and Workshops (IGCC), 2011 International, pages 1–8. IEEE,
2011.
[15] Johan Tordsson, Rubén S Montero, Rafael Moreno-Vozmediano, and Ignacio M Llorente. Cloud
brokering mechanisms for optimized placement of virtual machines across multiple providers.
Future Generation Computer Systems, 28(2):358–367, 2012.
Cloud Computing Resource Allocation Taxonomies 19
[16] Fabio López-Pires and Benjamín Barán. Multi-objective virtual machine placement with service
level agreement: A memetic algorithm approach. In Proceedings of the 2013 IEEE/ACM 6th
International Conference on Utility and Cloud Computing, pages 203–210. IEEE Computer
Society, 2013.
[17] Katsunori Sato, Masaki Samejima, and Norihisa Komoda. Dynamic optimization of virtual
machine placement by resource usage prediction. In Industrial Informatics (INDIN), 2013 11th
IEEE International Conference on, pages 86–91. IEEE, 2013.
[18] Kuan-yin Chen, Yang Xu, Kang Xi, and H Jonathan Chao. Intelligent virtual machine placement
for cost efficiency in geo-distributed cloud systems. In Communications (ICC), 2013 IEEE
International Conference on, pages 3498–3503. IEEE, 2013.
[19] Eyal Bin, Ofer Biran, Odellia Boni, Erez Hadad, Elliot K Kolodner, Yosef Moatti, and Dean H
Lorenz. Guaranteeing high availability goals for virtual machine placement. In Distributed
Computing Systems, 2011 31st International Conference on, pages 700–709. IEEE, 2011.
[20] Corentin Dupont, Giovanni Giuliani, Fabien Hermenier, Thomas Schulze, and Andrey Somov.
An energy aware framework for virtual machine placement in cloud federated data centres. In
Future Energy Systems: Where Energy, Computing and Communication Meet (e-Energy), 2012
Third International Conference on, pages 1–10. IEEE, 2012.
[21] Ching Chuen Teck Mark, Dusit Niyato, and Tham Chen-Khong. Evolutionary optimal virtual
machine placement and demand forecaster for cloud computing. In Advanced Information
Networking and Applications (AINA), 2011 IEEE International Conference on, pages 348–355.
IEEE, 2011.
[22] Zhe Huang and Danny HK Tsang. Sla guaranteed virtual machine consolidation for computing
clouds. In Communications (ICC), 2012 IEEE International Conference on, pages 1314–1319.
IEEE, 2012.
[23] Kuan Lu, Ramin Yahyapour, Philipp Wieder, Constantinos Kotsokalis, Edwin Yaqub, and Ali Im-
ran Jehangiri. Qos-aware vm placement in multi-domain service level agreements scenarios.
In Cloud Computing (CLOUD), 2013 IEEE Sixth International Conference on, pages 661–668.
IEEE, 2013.
[24] Kamal Bouyoucef, I Limam-Bedhiaf, and Omar Cherkaoui. Optimal allocation approach of
virtual servers in cloud computing. In Next Generation Internet (NGI), 2010 6th EURO-NF
Conference on, pages 1–6. IEEE, 2010.
[25] D. Espling, L. Larsson, W. Li, J. Tordsson, and E. Elmroth. Modeling and placement of cloud
services with internal structure. Cloud Computing, IEEE Transactions on, PP(99):1–1, 2014.
[26] Fabio López-Pires, Elias Melgarejo, and Benjamín Barán. Virtual machine placement. a multi-
objective approach. In Computing Conference (CLEI), 2013 XXXIX Latin American, pages 1–8.
IEEE, 2013.
[27] Fabio López-Pires and Benjamín Barán. A many-objective optimization framework for virtual-
ized datacenters. In Proceedings of the 2015 5th International Conference on Cloud Computing
and Service Science, pages 439–450, 2015.
[28] Jammily Ortigoza, Fabio López Pires, and Benjamín Barán. A taxonomy on dynamic environ-
ments for provider-oriented virtual machine placement. In Proceedings of the 2016 IEEE 4th
International Conference on Cloud Engineering. IEEE Computer Society, 2016.
[29] Wubin Li, Johan Tordsson, and Erik Elmroth. Modeling for dynamic cloud scheduling via
migration of virtual machines. In Cloud Computing Technology and Science (CloudCom), 2011
IEEE Third International Conference on, pages 163–171. IEEE, 2011.
[30] José Luis Lucas Simarro, Rafael Moreno-Vozmediano, Ruben S Montero, and Ignacio Martín
Llorente. Dynamic placement of virtual machines for cost optimization in multi-cloud environ-
ments. In High Performance Computing and Simulation (HPCS), 2011 International Conference
on, pages 1–7. IEEE, 2011.
20 Fabio López-Pires and Benjamín Barán
[31] Huynh Tu Dang and Fabien Hermenier. Higher SLA satisfaction in datacenters with continuous
VM placement constraints. In Proceedings of the 9th Workshop on Hot Topics in Dependable
Systems, page 1. ACM, 2013.
[32] Zhe Huang, Danny HK Tsang, and James She. A virtual machine consolidation framework
for mapreduce enabled computing clouds. In Proceedings of the 24th International Teletraffic
Congress, page 26. International Teletraffic Congress, 2012.
[33] Weiming Shi and Bo Hong. Towards profitable virtual machine placement in the data center.
In Utility and Cloud Computing (UCC), 2011 Fourth IEEE International Conference on, pages
138–145. IEEE, 2011.
[34] Benjamín Barán, Christian von Lücken, and Aldo Sotelo. Multi-objective pump scheduling
optimisation using evolutionary strategies. Advances in Engineering Software, 36(1):39–47,
2005.
[35] Carlos Coello Coello, Gary B Lamont, and David A Van Veldhuizen. Evolutionary algorithms
for solving multi-objective problems. Springer, 2007.
[36] Hadi Goudarzi and Massoud Pedram. Energy-efficient virtual machine replication and placement
in a cloud computing system. In Cloud Computing (CLOUD), 2012 IEEE 5th International
Conference on, pages 750–757. IEEE, 2012.
[37] Jing Tai Piao and Jun Yan. A network-aware virtual machine placement and migration approach
in cloud computing. In Grid and Cooperative Computing (GCC), 2010 9th International Con-
ference on, pages 87–92. IEEE, 2010.
[38] Wubin Li, Johan Tordsson, and Erik Elmroth. Virtual machine placement for predictable and
time-constrained peak loads. In Economics of Grids, Clouds, Systems, and Services, pages
120–134. Springer, 2012.
[39] Meng Sun, Weidong Gu, Xinchang Zhang, Huiling Shi, and Wei Zhang. A matrix transformation
algorithm for virtual machine placement in cloud. In Trust, Security and Privacy in Computing
and Communications (TrustCom), 2013 12th IEEE International Conference on, pages 1778–
1783. IEEE, 2013.
[40] Fei Song, Daochao Huang, Huachun Zhou, Hongke Zhang, and Ilsun You. An optimization-based
scheme for efficient virtual machine placement. International Journal of Parallel Programming,
42(5):853–872, 2014.
[41] Daniel S Dias and Luıs Henrique MK Costa. Online traffic-aware virtual machine placement in
data center networks. In Global Information Infrastructure and Networking Symposium (GIIS),
2012, pages 1–8. IEEE, 2012.
[42] Abhishek Gupta, Laxmikant V Kalé, Dejan Milojicic, Paolo Faraboschi, and Susanne M Balle.
Hpc-aware vm placement in infrastructure clouds. In Cloud Engineering (IC2E), 2013 IEEE
International Conference on, pages 11–20. IEEE, 2013.
[43] Jiankang Dong, Hongbo Wang, Xing Jin, Yangyang Li, Peng Zhang, and Shiduan Cheng. Virtual
machine placement for improving energy efficiency and network performance in iaas cloud. In
Distributed Computing Systems Workshops (ICDCSW), 2013 IEEE 33rd International Confer-
ence on, pages 238–243. IEEE, 2013.
[44] Hua-Jun Hong, De-Yu Chen, Chun-Ying Huang, Kuan-Ta Chen, and Cheng-Hsin Hsu. Qoe-
aware virtual machine placement for cloud games. In Network and Systems Support for Games
(NetGames), 2013 12th Annual Workshop on, pages 1–2. IEEE, 2013.
[45] Aissan Dalvandi, Mohan Gurusamy, and Kee Chaing Chua. Time-aware vm-placement and
routing with bandwidth guarantees in green cloud data centers. In Cloud Computing Technology
and Science (CloudCom), 2013 IEEE 5th International Conference on, volume 1, pages 212–217.
IEEE, 2013.
Cloud Computing Resource Allocation Taxonomies 21
[46] Wenting Wang, Haopeng Chen, and Xi Chen. An availability-aware virtual machine placement
approach for dynamic scaling of cloud applications. In Ubiquitous Intelligence & Computing
and 9th International Conference on Autonomic & Trusted Computing (UIC/ATC), 2012 9th
International Conference on, pages 509–516. IEEE, 2012.
[47] Nicolò Maria Calcavecchia, Ofer Biran, Erez Hadad, and Yosef Moatti. VM placement strategies
for cloud scenarios. In Cloud Computing (CLOUD), 2012 IEEE 5th International Conference
on, pages 852–859. IEEE, 2012.
[48] Zhibo Cao and Shoubin Dong. An energy-aware heuristic framework for virtual machine con-
solidation in cloud computing. The Journal of Supercomputing, pages 1–23, 2014.
[49] Maolin Tang and Shenchen Pan. A hybrid genetic algorithm for the energy-efficient virtual
machine placement problem in data centers. Neural Processing Letters, pages 1–11, 2014.
[50] Konstantinos Tsakalozos, Mema Roussopoulos, and Alex Delis. Vm placement in non-
homogeneous iaas-clouds. In Service-Oriented Computing, pages 172–187. Springer, 2011.
[51] Jing Xu and Jose AB Fortes. Multi-objective virtual machine placement in virtualized data center
environments. In Green Computing and Communications (GreenCom), 2010 IEEE/ACM Int’l
Conference on & Int’l Conference on Cyber, Physical and Social Computing (CPSCom), pages
179–188. IEEE, 2010.
[52] Soichi Shigeta, Hiroyuki Yamashima, Tsunehisa Doi, Tsutomu Kawai, and Keisuke Fukui.
Design and implementation of a multi-objective optimization mechanism for virtual machine
placement in cloud computing data center. In Cloud Computing, pages 21–31. Springer, 2013.
[53] Amol C Adamuthe, Rupali M Pandharpatte, and Gopakumaran T Thampi. Multiobjective virtual
machine placement in cloud environment. In Cloud & Ubiquitous Computing & Emerging
Technologies (CUBE), 2013 International Conference on, pages 8–13. IEEE, 2013.
[54] Yongqiang Gao, Haibing Guan, Zhengwei Qi, Yang Hou, and Liang Liu. A multi-objective
ant colony system algorithm for virtual machine placement in cloud computing. Journal of
Computer and System Sciences, 79(8):1230–1242, 2013.
[55] Jan-Jan Wu, Pangfeng Liu, and Jyun-Shiung Yang. Workload characteristics-aware virtual ma-
chine consolidation algorithms. In Proceedings of the 2012 IEEE 4th International Conference
on Cloud Computing Technology and Science (CloudCom), pages 42–49. IEEE Computer So-
ciety, 2012.
[56] Weiwei Fang, Xiangmin Liang, Shengxin Li, Luca Chiaraviglio, and Naixue Xiong. Vmplanner:
Optimizing virtual machine placement and traffic flow routing to reduce network power costs in
cloud data centers. Computer Networks, 57(1):179–196, 2013.
[57] Jixiang Cheng, Gary G. Yen, and Gexiang Zhang. A many-objective evolutionary algorithm
based on directional diversity and favorable convergence. In Systems, Man and Cybernetics
(SMC), 2014 IEEE International Conference on, pages 2415–2420, Oct 2014.
[58] Marco Farina and Paolo Amato. On the optimal solution definition for many-criteria optimization
problems. In Proceedings of the NAFIPS-FLINT international conference, pages 233–238, 2002.
[59] Kalyanmoy Deb, Ankur Sinha, and Saku Kukkonen. Multi-objective test problems, linkages,
and evolutionary methodologies. In Proceedings of the 8th annual conference on Genetic and
evolutionary computation, pages 1141–1148. ACM, 2006.
[60] Christian von Lücken, Benjamín Barán, and Carlos Brizuela. A survey on multi-objective evolu-
tionary algorithms for many-objective problems. Computational Optimization and Applications,
pages 1–50, 2014.
[61] Luiz André Barroso and Urs Hölzle. The case for energy-proportional computing. IEEE com-
puter, 40(12):33–37, 2007.
[62] Jiankang Dong, Xing Jin, Hongbo Wang, Yangyang Li, Peng Zhang, and Shiduan Cheng. Energy-
saving virtual machine placement in cloud data centers. In Cluster, Cloud and Grid Computing
(CCGrid), 2013 13th IEEE/ACM International Symposium on, pages 618–624. IEEE, 2013.
22 Fabio López-Pires and Benjamín Barán
[63] Shao-Heng Wang, Patrick P-W Huang, Charles H-P Wen, and Li-Chun Wang. Eqvmp: Energy-
efficient and qos-aware virtual machine placement for software defined datacenter networks.
In Information Networking (ICOIN), 2014 International Conference on, pages 220–225. IEEE,
2014.
[64] Nick McKeown, Tom Anderson, Hari Balakrishnan, Guru Parulkar, Larry Peterson, Jennifer
Rexford, Scott Shenker, and Jonathan Turner. Openflow: enabling innovation in campus net-
works. ACM SIGCOMM Computer Communication Review, 38(2):69–74, 2008.
[65] Orna Agmon Ben-Yehuda, Muli Ben-Yehuda, Assaf Schuster, and Dan Tsafrir. Deconstructing
Amazon EC2 spot instance pricing. ACM Transactions on Economics and Computation, 1(3):16,
2013.
[66] Wubin Li, Petter Svärd, Johan Tordsson, and Erik Elmroth. Cost-optimal cloud service placement
under dynamic pricing schemes. In Proceedings of the 2013 IEEE/ACM 6th International
Conference on Utility and Cloud Computing, pages 187–194. IEEE Computer Society, 2013.
[67] Kangkang Li, Jie Wu, and Adam Blaisse. Elasticity-aware virtual machine placement for cloud
datacenters. In Cloud Networking (CloudNet), 2013 IEEE 2nd International Conference on,
pages 99–107. IEEE, 2013.
[68] Mayank Mishra and Anirudha Sahoo. On theory of vm placement: Anomalies in existing
methodologies and their mitigation using a novel vector based approach. In Cloud Computing
(CLOUD), 2011 IEEE International Conference on, pages 275–282. IEEE, 2011.
[69] Tiago Ferreto, César AF De Rose, and Hans-Ulrich Heiss. Maximum migration time guaran-
tees in dynamic server consolidation for virtualized data centers. In Euro-Par 2011 Parallel
Processing, pages 443–454. Springer, 2011.
[70] Diego Ihara, Fabio López-Pires, and Benjamín Barán. Many-objective virtual machine placement
for dynamic environments. In Proceedings of the 2015 IEEE/ACM 8th International Conference
on Utility and Cloud Computing. IEEE Computer Society, 2015.
[71] Mohammad Mehedi Hassan, M Shamim Hossain, AM Jehad Sarkar, and Eui-Nam Huh. Co-
operative game-based distributed resource allocation in horizontal dynamic cloud federation
platform. Information Systems Frontiers, 16(4):523–542, 2014.
... The VMP problem could be formulated as both online and offline optimization problems [2]. In order to improve the quality of solutions given by online algorithms, the VMP problem could also be formulated as a twophase optimization scheme, which combines advantages of online and offline formulations for IaaS environments, as previously demonstrated in [1]. ...
... As presented in Column 2 of Table I, the VMPr phase may be considered with centralized (C) or distributed (D) approaches when considering a two-phase optimization scheme for VMP problems. In a centralized decision approach (C), the optimization is globally performed, evaluating the placement of all allocated VMs, while a distributed approach (D) partially reconfigures VMs allocated in one isolated PM [2]. ...
... It is worth mentioning that power consumption management has become a crucial study issue in the provideroriented VMP literature [2]. In this context, A32 algorithm shows a promising future for green cloud datacenters efficiently managing VMP decisions. ...
... These criteria can even change from one period of time to another, which implies a variety of possible environments, formulations and different objectives to be considered for optimization of the VMP problem. In this context, nearly 60 different objective functions were identified in the VMP literature [93,94,96]. ...
... Additionally, determining which solution to keep and which to discard in order to converge toward the Pareto set is still a relevant issue to be addressed [45], making more difficult to solve MaOPs. Clearly, existing difficulties in solving MaOPs explain why it is still considered an unexplored domain in resource management of cloud computing datacenters [58] and no many-objective formulations have already been proposed for the VMP problem in the specialized literature before this doctoral thesis [94,96]. ...
... 1. Taxonomies of VMP problems for cloud computing (Objective 1) [10,94,96]. ...
... Next, the authors extended their previous work in [47] with novel taxonomies, to present a detailed view of the existing approaches as well as several possible research opportunities to further advance in this research area. The taxonomies presented in [48] could guide interested readers to: (1) understand different possible environments where a VMP problem could be studied, considering both provider and broker perspectives in different deployment architectures (see Section 2), (2) identify existing approaches for the formulation and resolution of the VMP as an optimization problem (see Section 3) and (3) present a detailed view of the VMP problem, identifying research opportunities to further advance in cloud computing resource allocation areas (see Section 4). ...
... Based on the taxonomies presented in [47] and [48], this chapter summarizes relevant concepts related to the VMP problem, including formal definitions that could help the research community to avoid terminological ambiguity as well as to follow common terminology and concepts. The remainder of this chapter is organized in the following way: Section 2 presents a VMP problem environment taxonomy for the classification of related articles by: (1) orientation, (2) deployment architecture and (3) types of formulation. ...
... Depending on the particular environment where a VMP problem will be studied, several different considerations should be taken into account before proposing a particular formulation or technique for the resolution of the considered VMP problem. For a complete understanding of the possible environments where a VMP problem could be studied, considering both provider and broker orientations in four different deployment architectures for online and offline formulations, Figure 2.1 presents the taxonomy proposed in [48] described in this section, including relevant references from the studied VMP literature [46]. ...
Full-text available
Chapter
Cloud computing datacenters dynamically provide millions of virtual machines in real-world cloud computing environments. A large number of research challenges have to be addressed toward an efficient resource management of these cloud computing infrastructures. In the resource allocation field, Virtual Machine Placement (VMP) is one of the most studied problems with several possible formulations and a large number of existing optimization criteria, considering solutions with high economical and ecological impact. Based on systematic reviews of the VMP literature, a taxonomy of VMP problem environments is presented to understand different possible environments where a VMP problem could be considered, from both provider and broker perspectives in different deployment architectures. Additionally, another taxonomy for VMP problems is presented to identify existing approaches for the formulation and resolution of the VMP as an optimization problem. Finally a detailed view of the VMP problem is presented, identifying research opportunities to further advance in cloud computing resource allocation areas.
... 14 However, López-Pires and Barán suggest that this is easier said than done as the efficient resource management of cloud infrastructures is highly challenging. 15 For Lee, most capacity planning and management area studies focus on micro-level scheduling such as dynamic resource allocation and prioritisation of computing jobs. 13 Widely used resource management methods like AWS's Auto Scaling and Azure's Autoscaling Application Block are both reactive; they are also provided by companies in a fox-guardingthe-henhouse kind of way, which is to say, their profits are based on usage, so are they really the best companies to provide applications that recommend use limitations? ...
... AMA0237_PEARSON.indd15 10/02/21 12:48 PM ...
Full-text available
Article
Capacity management is the process of optimising production in line with fluctuating demand for products and services in order to reduce wasted capacity. Simply put, it aims to make optimal use of essential resources while minimising the use of non-essential resources. To this end, the key is to balance the right number of users and the right performance at peak usage to ensure a great end-user experience. This paper explores a capacity planning solution that can predict upcoming costs with advanced predictive analytics and forward-thinking what-if scenario modelling that can produce a healthy return on investment as well as help companies go green.
... The identified challenges for considered VMP problems [3] are presented on the particular context of 5G services, mainly taking into account ML techniques for addressing relevant decision making on cloud computing infrastructure operations (e.g. when or under what circumstances a VMPr phase should be triggered?). Additionally, network management challenges based on Software-Defined Networking (SDN) are also analyzed from the perspective of VMP problems, where ML techniques may result on a promising approach to support these type of operational decisions. ...
... reconfiguration time), where γ may vary according to the maximum amount of RAM to be migrated. Figure 1 presents the described twophase optimization scheme, considering β = 2, from t = 2 to t = 4 and γ = 1, from t = 4 to t = 5. References VMPr Triggering [9,10,11,12,13,14,15] Periodically [16,17,18] Threshold-based [4] Prediction-based It is important to notice that a large number of possible objective functions F(x,t) and constraints e(x,t) could be considered for a VMP problem formulation, according to provider preferences [3,19]. ...
... Resource allocation with fairness in cloud computing has been widely considered to be the most challenging issue. Despite other distributed systems, cloud computing is particularly recognized in the heterogeneity of resources and servers [1]. In other words, a cloud data center is likely to be established by different servers, including diverse configurations in terms of resources, such as memory, processing, and disk storage [2]. ...
... The efficient and accurate assessment of cloud-based infrastructure is essential for guaranteeing both business continuity and uninterrupted services for computing jobs [21]. However, efficient resource management of the cloud infrastructures is challenging [22]. In the capacity planning and management area, most studies focus on micro-level scheduling such as dynamic resource allocation and prioritization of computing jobs. ...
Full-text available
Article
While the rapid growth of cloud computing is driven by the surge of big data, the Internet of Things, and social media applications, an evaluation and investment decision for cloud computing has been challenging for corporate managers due to a lack of proper decision models. This paper attempts to identify critical variables for making a cloud capacity decision from a corporate customer’s perspective and develops a base mathematical model to aid in a hybrid cloud investment decision under probabilistic computing demands. The identification of the critical variables provides a means by which a corporate customer can effectively evaluate various cloud capacity investment opportunities. Critical variables included in this model are an actual computing demand, the amount of private cloud capacity purchased, the purchase cost of the private cloud capacity, the price of the public cloud, and the default downtime loss/penalty cost. Extending the base model developed, this paper also takes into consideration the interoperability cost incurred in cloud bursting to the public cloud and derives the optimal investment. The interoperable cloud systems require time and investment by the users and/or cloud providers and there exists a diminishing return on the investment. Hence, the relationship between the interoperable cloud investment and return on investment is also investigated.
... There are various examples in the literature discussing resource allocation in cloud computing environments. The best ways for distributing resources among users with varying demands have been investigated extensively in [3]. As part of this work, different quality metrics were introduced and among them, fairness has been identified as an important and challenging issue , attracting more attention in recent years by the research community. ...
Chapter
Cloud Computing provides on-demand, flexible, ubiquitous resources for clients in a virtualized environment using huge number of virtual machines (VMs). Cloud data centers don’t utilize their resources fully which leads into a underutilization of resources. Virtualization offers a few exceptional highlights for cloud suppliers like saving of power consumption, load adjusting, and adaptation to internal failure, resource multiplexing. However, for improving energy proficiency and resource utilization, various strategies have been introduced such as server consolidation and different resource structuring. Among all, Virtual Machine Placement (VMP) is the most vital strides in server consolidation. Virtual Machine Placement (VMP) is an efficient mapping of VMs to Physical Machines (PMs). VMP issues go about as a non-deterministic polynomial-time hard (NP-difficult) issue and metaheuristics strategies are widely used to solve these issues with enhancing boundaries of power utilization, QoS, resource usage, etc. This paper presents an extensive review of Metaheuristics models to deal with VMP in the cloud environment.
Full-text available
Article
The prosperity of cloud technology generates expanding numerous real-time applications runs online. Meantime, real time tasks scheduling is an important criteria for cloud service provider to manage its Quality of Service (QoS) because all customer’s wishes gratifying resource allotment in cloud. Cloud Service Provider (CSP) constitute service level agreement with the customers before the provision of resources. Allocating the resource according to the customer need and utilizing the systems efficiently are the major concern of CSP to increase the profit in their business. For this purpose Hungarian optimization technique is used and it is a standard technique which provides load balanced allocation of resources to the task. Still it cannot be used directly for cloud Virtual Machine (VM) allocation, because load balancing will not give better makespan and the standard Hungarian method is not suitable for unbalanced cost matrix. In this paper efficient resource allocation method called Cluster Cost Matrix – Hungarian (CCM-H) algorithm is proposed to optimize the performance. Algorithm consists of two phases. In first phase algorithm calculates the weighted values of tasks and based on the value tasks are clustered to convert the unbalanced cost matrix to balanced cost matrix. In second phase, according to the balanced cost matrix VM allocation is performed using Hungarian optimization technique. The metrics used for the performance analysis are makespan and utilization factor. The proposed CCM-H algorithm is compared with various existing and standard algorithms called First Come First Serve (FCFS), Min–Min based iterative Hungarian, FCFS based iterative algorithm, Max–min based iterative algorithm Laha and Gupta (Comput Ind Eng, 2016), Group based algorithms Lu et al. (IEEE Trans Wirel Commun, 2017) and normal Hungarianalgorithm, with bench mark dataset Braun (2015) and synthetic dataset which is created with random number generation function. Output shows that how the proposed method outperforms all the existing models.
Full-text available
Conference Paper
Cloud computing datacenters provide millions of virtual machines in actual cloud markets. In this context, Virtual Machine Placement (VMP) is one of the most challenging problems in cloud infrastructure management, considering the large number of possible optimization criteria and different formulations that could be studied. Considering the on-demand model of cloud computing, the VMP problem should be solved dynamically to efficiently attend typical workload of modern applications. This work proposes a taxonomy in order to understand possible challenges for Cloud Service Providers (CSPs) in dynamic environments, based on the most relevant dynamic parameters studied so far in the VMP literature. Based on the proposed taxonomy, several unexplored environments have been identified. To further study those research opportunities, sample workload traces for each particular environment are required; therefore, basic examples illustrate a preliminary work on dynamic workload trace generation.
Full-text available
Conference Paper
This paper presents for the first time a formulation of the Virtual Machine Placement as a Many-Objective problem (MaVMP), considering the simultaneous optimization of the following five objective functions for dynamic environments: (1) power consumption, (2) inter-VM network traffic, (3) economical revenue, (4) number of VM migrations and (5) network traffic overhead for VM migrations. To solve the formulated MaVMP problem, a novel Memetic Algorithm is proposed. As a potentially large number of feasible solutions at any time is one of the challenges of MaVMP, five selection strategies are evaluated in order to automatically select one solution at each time. The proposed algorithm with the considered selection strategies were evaluated in two different scenarios.
Full-text available
Conference Paper
The process of selecting which virtual machines should be located (i.e. executed) at each physical machine of a datacenter is commonly known as Virtual Machine Placement (VMP). This work presents a general manyobjective optimization framework that is able to consider as many objective functions as needed when solving the VMP problem in a pure multi-objective context. As an example of utilization of the proposed framework, for the first time a formulation of the many-objective VMP problem (MaVMP) is proposed, considering the simultaneous optimization of the following five objective functions: (1) power consumption, (2) network traffic, (3) economical revenue, (4) quality of service and (5) network load balancing. To solve the formulated many-objective VMP problem, an interactive memetic algorithm is proposed. Simulations prove the correctness of the proposed algorithm and its effectiveness converging to a treatable number of solutions in different experimental scenarios.
Full-text available
Conference Paper
Cloud computing datacenters dynamically provide millions of virtual machines (VMs) in actual cloud markets. In this context, Virtual Machine Placement (VMP) is one of the most challenging problems in cloud infrastructure management, considering the large number of possible optimization criteria and different formulations that could be studied. VMP literature include relevant research topics such as energy efficiency, Service Level Agreement (SLA), Quality of Service (QoS), cloud service pricing schemes and carbon dioxide emissions; all of them with high economical and ecological impact. This work classifies an extensive up-to-date survey of the most relevant VMP literature proposing a novel taxonomy in order to identify research opportunities and define a general vision on this research area.
Full-text available
Article
Server consolidation using virtualization technology has become an important technology to improve the energy efficiency of data centers. Virtual machine placement is the key in the server consolidation technology. In the past few years, many approaches to the virtual machine placement have been proposed. However, existing virtual machine placement approaches consider the energy consumption by physical machines only, but do not consider the energy consumption in communication network, in a data center. However, the energy consumption in the communication network in a data center is not trivial, and therefore should be considered in the virtual machine placement. In our preliminary research, we have proposed a genetic algorithm for a new virtual machine placement problem that considers the energy consumption in both physical machines and the communication network in a data center. Aiming at improving the performance and efficiency of the genetic algorithm, this paper presents a hybrid genetic algorithm for the energy-efficient virtual machine placement problem. Experimental results show that the hybrid genetic algorithm significantly outperforms the original genetic algorithm, and that the hybrid genetic algorithm is scalable.
Conference Paper
In this paper, we present the methodologies used in existing literature for Virtual Machine (VM) placement, load balancing and server consolidation in a data center environment. While the methodologies may seem fine on the surface, certain drawbacks and anomalies can be uncovered when they are analyzed deeper. We point out those anomalies and drawbacks in the existing literature and explain what are the root causes of such anomalies. Then we propose a novel methodology based on vector arithmetic which not only addresses those anomalies but also leads to some interesting theories and algorithms to tackle the above mentioned three functionalities required in managing resources of data centers. We believe that with a strong mathematical base, our methodology has the potential to become the foundation of future models and algorithms in this research area. There are few research work reported in the literature for VM placement. Those methods might look fine at a glance, but a deeper scrutiny can expose various anomalies and drawbacks which might affect the performance of the system. Most of them devise a metric, which is a function of resource utilizations of individual resource types. They use this metric for placement and migration of VMs as well as for load balancing and consolidation of servers. In this paper, we present various methodologies used in the literature for VM placement, server load balancing and server consolidation and point out the drawbacks and anomalies in those methodologies and discuss the root cause of such anomalies. Then we propose a novel methodology based on vector arithmetic which not only addresses those anomalies but also leads to some inter- esting theories and algorithms to tackle the above mentioned three functionalities required in managing resources of data centers. We believe that with a strong mathematical base, our methodology has the potential to become the foundation of future models and algorithms in this research area.
Conference Paper
Cloud computing is becoming a popular way of supplying and using computing resources. A cloud-computing data center is equipped with a large number of physical resources and must manage an even larger number of virtual machines (VMs). The center’s VM placement strategy affects the utilization of physical resources, and consequently, it influences operational costs. Our goal is to develop a multi-objective optimization mechanism for VM placement that satisfies various constraints and results in the lowest operational cost. The number of possible combinations of VMs and hosts can be extremely large. For the mechanism to be practical, the number of possible combinations must be reduced. We reduced computational overheads by classifying VM hosts into a relatively small number of equivalent sets. Simulation results show that expected operational costs can be significantly reduced by applying the proposed mechanism.
Conference Paper
We present an approach to optimal virtual machine placement within datacenters for predicable and time-constrained load peaks. A method for optimal load balancing is developed, based on binary integer programming. For tradeoffs between quality of solution and computation time, we also introduce methods to pre-process the optimization problem before solving it. Upper bound based optimizations are used to reduce the time required to compute a final solution, enabling larger problems to be solved. For further scalability, we also present three approximation algorithms, based on heuristics and/or greedy formulations. The proposed algorithms are evaluated through simulations based on synthetic data sets. The evaluation suggests that our algorithms are feasible, and that these can be combined to achieve desired tradeoffs between quality of solution and execution time.
Conference Paper
The rapid growing demand for computational power by modern applications in Cloud Computing has led to the creation of large-scale data centers. Such data centers consume huge amount of electrical energy, emitting a great deal of CO2 in our environment. One effective way so as to reduce energy consumption is consolidating the vms using dynamic migration on minimum number of nodes and switching idle nodes to the sleep or hibernate modes which can cause better resource usage. But, due to the varying workloads in applications, consolidating the vms can cause SLA violation specially in cases that resource usage exceeds upper threshold defined for such nodes including consolidated vms. On the one hand, due to the necessity of providing the high QoS for customers which is negotiated in terms of SLA in cloud computing and on the other hand, importance of energy consumption as a critical problem, there is the necessity in addressing performance-energy trade off - the minimization of energy consumption, at the same time to take into account QoS. In this paper, related works regarding energy efficient datacentres have been conducted and in the end they have been compared with each other.
Article
Virtual machine placement is the process of mapping virtual machines to available physical hosts within a datacenter or on a remote datacenter in a cloud federation. Normally, service owners cannot influence the placement of service components beyond choosing datacenter provider and deployment zone at that provider. For some services, however, this lack of influence is a hindrance to cloud adoption. For example, services that require specific geographical deployment (due e.g. to legislation), or require redundancy by avoiding co-location placement of critical components. We present an approach for service owners to influence placement of their service components by explicitly specifying service structure, component relationships, and placement constraints between components. We show how the structure and constraints can be expressed and subsequently formulated as constraints that can be used in placement of virtual machines in the cloud. We use an integer linear programming scheduling approach to illustrate the approach, show the corresponding mathematical formulation of the model, and evaluate it using a large set of simulated input. Our experimental evaluation confirms the feasibility of the model and shows how varying amounts of placement constraints and data center background load affects the possibility for a solver to find a solution satisfying all constraints within a certain timeframe. Our experiments indicate that the number of constraints affects the ability of finding a solution to a higher degree than background load, and that for a high number of hosts with low capacity, component affinity is the dominating factor affecting the possibility to find a solution.