Towards Optimized Security-aware (O-Sec) VM Placement Algorithms
Motlatsi Isaac Thulo and J. H. P. Eloff
Cybersecurity and Big Data Science Research Group, Department of Computer Science, University of Pretoria,
Pretoria, South Africa
Keywords:
Cloud Computing, VM Placement Algorithm, Security-aware VM Placement.
Abstract:
Cloud computing is a technology that takes advantage of virtualization. Through virtualization, Virtual Ma-
chines (VMs) within the same host machine share physical resources. Cloud service providers (CSP) take
advantage of virtualization by providing on-demand computing resources through the use of the Internet. In
order to provide good Quality of Service (QoS) and to lower costs, CSPs need to optimize the cloud environ-
ment. This optimization can be achieved by the strategic placement of Virtual Machines (VMs) in cloud archi-
tecture, usually through VM placement algorithms. Despite these efforts, there are some remaining problems
that need to be addressed. Amongst these are threats introduced by the cloud’s architectural vulnerabilities.
This paper, therefore, focuses on evaluating currently available VM placement algorithms. The objective is
to identify VM placement algorithms that show potential to be further augmented with security features or
that can be improved from a security perspective. Future work will investigate how these algorithms can be
adapted to be security-aware.
1 INTRODUCTION
Cloud computing is a rapidly growing technology
which takes advantage of virtualization as an underly-
ing tool. Through virtualization, multiple instances of
operating systems run on a single Physical Machine
(PM) and share physical resources like CPU, mem-
ory and storage (Biran et al., 2012). Cloud Service
Providers (CSPs) make use of this technology by pro-
viding an on-demand computing resources through
the use of the Internet. This allows organizations
to reduce their costs of maintenance and administra-
tion for their datacentres (Dong et al., 2015). The
types of cloud computing models offered by CSPs are
Platform as a Service (PaaS), Software as a Service
(SaaS), and Infrastructure as a Service (IaaS) (Al-Haj
et al., 2013). The ability of IaaS to provision VMs
allows it to offer highest level of flexibility and scal-
ability (Yuchi and Shetty, 2015). However, according
to article (Lindemann, 2015) there are some failures
in the sufficient protection of IaaS cloud resources.
These failures are brought by, for example, the re-
stricted view of the VMs by Intrusion Detection Sys-
tem (IDS) and limited anomaly detection techniques
within the cloud environment. This therefore allows
malicious activities, such as unauthorized inter-VM
communication, that affect cloud computing users.
CSPs are expected to implement countermeasures for
minimizing the risk of such malicious activities but
they are expected, at the same time, to provide good
service at a reasonable cost.
CSPs need to provide good Quality of Service
(QoS) to satisfy paying customers at lower costs.
There are some ongoing research efforts on optimiza-
tion of the cloud architecture to reduce costs with-
out violating the Service Level Agreements (SLA)
(Gao and Tang, 2013), (Xu and Fortes, 2011), (Kuo
et al., 2015), (Ohta, 2013), (Meng et al., 2010). Ac-
cording to the last mentioned references, the strategic
placement of VMs within the cloud architecture is the
key to ensuring good QoS and reduced costs. These
references, amongst others, focus on objectives such
as energy-consumption minimization (Dong et al.,
2015), (Vu and Hwang, 2014), resource-utilization
(Ohta, 2013), (Zaouch and Benabbou, 2015), work-
load or network-management (Ilkhechi et al., 2015),
(Vu and Hwang, 2014), overhead minimization (An-
dreolini et al., 2009) and time to complete job (Li
et al., 2015). The said objectives of the last mentioned
references all contribute towards one goal, which is
optimization of the cloud environment. But they do
not, however, minimize the security threats imposed
in an IaaS cloud.
In addition to optimization, security also needs to
be taken into consideration in the implementation of
VM placement algorithms. IaaS cloud infrastructure
Thulo, M. and Eloff, J.
Towards Optimized Security-aware (O-Sec) VM Placement Algorithms.
DOI: 10.5220/0006206504110422
In Proceedings of the 3rd International Conference on Information Systems Security and Privacy (ICISSP 2017), pages 411-422
ISBN: 978-989-758-209-7
Copyright
c
 2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
411

has architectural vulnerabilities such as virtual ma-
chine escape (Venkata et al., 2004), and data breaches
(Alani, 2014). These vulnerabilities in cloud infras-
tructure enable some cloud specific threats. These
threats take advantage of the fact that VMs share
physical resources (Afoulki et al., 2011). Accord-
ing to the article (Yuchi and Shetty, 2015), the com-
promised VMs within the cloud architecture use the
shared resources as channels to disrupt, corrupt or
spy on other VMs within the architecture. To address
this, there are some ongoing security related research
efforts in cloud computing. Amongst other security
related efforts are those in articles (Li et al., 2012),
(Yuchi and Shetty, 2015), (Afoulki et al., 2011),
(Caron et al., 2013), (Al-Haj et al., 2013). The ref-
erenced articles incorporate security objectives for the
strategic placement of VMs. Some of these, for exam-
ple, focus on the creation of security groups for VMs
based on the similarities of incoming traffic (Al-Haj
et al., 2013). Most of the current work on address-
ing security objectives for IaaS cloud however ignore
QoS and cost. It is therefore important to further con-
duct research into the notion of Optimized Security-
aware VM placement algorithms (O-Sec VM Place-
ment). This notion includes security, QoS as well as
cost optimization objectives.
In this paper, currently available VM placement
algorithms are studied. For each of the algorithms,
QoS, cost and security objectives are evaluated. The
next section discusses related work. Section 3 lists
and discusses the most important currently available
VM placement algorithms and then formulates the
evaluation criteria. The criteria are used to evaluate
the VM placement algorithms. Section 4 provides
suggestions for future work.
2 BACKGROUND AND RELATED
WORK
2.1 VM Placement Algorithm
The concept of VM placement algorithms originates
from the consolidation of Operating Systems (OSs)
within the same PM to avoid application compatibil-
ity issues (Bobroff et al., 2007). This consolidation of
OSs within the same PM also increases the utilization
of resources to lower running costs. In around 2007,
researchers found the need to use algorithms to auto-
matically consolidate these OSs (or VMs in this pa-
pers context) and allow dynamic placement into mul-
tiple PMs (Bobroff et al., 2007), (Kochut and Beaty,
2007). This concept is adopted by cloud computing
and the VM placement algorithms are used to strate-
gically allocate VMs to available PMs. The strate-
gic placement of VMs is based on client requirements
and aims to achieve particular objectives (Meng et al.,
2010), (Ohta, 2013), (Xu and Fortes, 2011), (Lin
et al., 2011). Amongst other objectives are energy
consumption minimization (Lin et al., 2011),(Yang
et al., 2012), load balancing (Zaouch and Benabbou,
2015), (Andreolini et al., 2009), traffic cost minimiza-
tion (Biran et al., 2012), (Meng et al., 2010), (Vu and
Hwang, 2014) and security-aware placement (Afoulki
et al., 2011), (Al-Haj et al., 2013), (Caron et al.,
2013).
In consideration of the optimization objectives de-
liberated in the currently available VM placement al-
gorithms and also taking into consideration the ob-
jectives of this paper, it is important to categorize the
placement algorithms into three broad categories. The
following categories will be used in the remainder of
this discussion: 1) VM placement algorithms that fo-
cus on optimizing QoS, 2) VM placement algorithms
that focus on cost reduction, and 3) VM placement
algorithms that focus on security-aware placement.
The next subsections will discuss the three categories,
with examples of the VM placement algorithms from
the most cited papers for each category. Due to space
constraints in the paper, only the first subsection that
discusses the first category includes activity diagrams
for the discussed VM placement.
2.2 VM Placement Algorithms Focusing
on QoS
The main aim of strategic placement of VMs within
the cloud architecture is to utilize the physical re-
source usage without violating SLA. According to
studies (Biran et al., 2012), (Zaouch and Benabbou,
2015), (Xu and Fortes, 2011) this strategic place-
ment of VMs strives to avoid performance degrada-
tion which can be caused by shared resources in the
cloud. Some authors, (Zaouch and Benabbou, 2015),
focus on the even distribution of VMs across the cloud
architecture to utilize resources and therefore provide
good QoS. Xu and Fortes (Xu and Fortes, 2011) on
the other hand, strive to avoid performance degrada-
tions that are caused by heat imbalances in cloud ar-
chitectures.
Among the mostly cited studies is (Biran et al.,
2012), whereby the authors focus on complex but
effective VM placement algorithms. This study fo-
cuses on both the physical resource and network con-
straints. It aims at implementing the VM placement
algorithm that is capable of absorbing the varying
traffic demands in cloud networks. The quadratic na-
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
412

ture of the proposed solutions brought by pairs of
communicating VMs, and the involvement of multi-
ple constraints, makes the implementation NP-Hard
and complex. The authors therefore propose heuristic
algorithms to solve the problem. These algorithms,
as shown in Figure 1, first cluster VMs into Con-
nected Components (CCs) where a connected com-
ponent represent a number of VMs communicating to
each other. They then iteratively place the CCs on
different levels of a tree network starting at the root
followed by the next levels of the tree, denoted by
nextLevel in Figure 1. If the nextLevel is the lowest
level of the tree then individual VMs are placed on
PMs, otherwise CCs are placed on switches, called
Virtual Hosts (VHs). If the aggregated resource re-
quirements of a CC
d
( for d = 1 to n) exceed the ag-
gregated capacities of V H
z
(for z = 1 to m), the CC
d
is split among the subtrees of the root. The iterative
placement of CCs on VHs result in communicating
VMs being placed into PMs that are in close proxim-
ity to each other. This minimises the total traffic costs
within the architecture. The aggregated multiple VM
placement algorithms which originates from the work
by (Biran et al., 2012) are interpreted by authors of
this paper using one single activity diagram as shown
in Figure 1.
Figure 1: MCRVMP.
The problem with their solution (Biran et al.,
2012) is that it works only with small to medium
sized cloud architectures. In summary, the salient key
points of this algorithm, from the perspective of the
study in hand, are its two properties: 1) The algo-
rithms is recursive, that is, it exploits the depth of
the tree network structure by solving the placement
of CCs or VMs on one level of the tree at the time. 2)
The algorithm adopts a two-phase approach, it firstly
places the CCs on the network and secondly expands
the CCs to place the actual VMs on the PMs. The two
properties together help the algorithm to place those
VMs with high communication cost between them in
the same PM or atleast in those that are in close prox-
imity to each other. This reduces the overall traffic
cost within the network structure and therefore min-
imises possibility of traffic bursts.
The other similar and mostly cited paper is (Meng
et al., 2010), also focusing on consumption of net-
work resources. The proposed VM placement algo-
rithms in the last mentioned article also strive to place
the pairs of communicating VMs close to each other
to reduce traffic costs and time to complete job. Vu
et al. (Vu and Hwang, 2014) also focus on consump-
tion of network resources, but they also include en-
ergy consumption minimization as an additional op-
timization objective. These studies collectively con-
sider the concept of connected components (CCs) that
share data (data in-transit). They however, do not fo-
cus on the security risks of that data, which are impor-
tant in the virtualized environment where resources
are shared. Security risks that are not implemented in
these cited papers (Biran et al., 2012), (Meng et al.,
2010), (Vu and Hwang, 2014) include, amongst oth-
ers the following:
• Communications within the cloud architecture use
shared network resources. This means that VMs
belonging to malicious users can intercept com-
munication between sibling VMs. This can be
achieved by either sniffing or spoofing IP ad-
dresses to redirect packets (Wu et al., 2010).
Sniffing, on one hand, is possible if the virtual net-
work is configured in bridge mode where a virtual
bridge acts as hub and broadcasts communication.
Spoofing, on the other hand, is possible if virtual
network is configured in route mode and a virtual
switch is used to direct packets to specific virtual
interfaces.
• In clustered CCs, some communicating VMs re-
side in either the same PM or those in close prox-
imity to each other. This makes it easier to iden-
tify and target a cluster of VMs that are close to
each other as opposed to identifying communi-
cating VMs that are in different locations. The
Towards Optimized Security-aware (O-Sec) VM Placement Algorithms
413

denial-of-services, which compromises availabil-
ity of VMs, is simple in this case as there are fewer
PMs to target.
2.3 VM Placement Focusing on Cost
Virtualization in cloud computing enables consolida-
tion of VMs within the PMs and reduces the costs of
having to buy servers for every application. In ad-
dition to this, the strategic placement of VMs within
the cloud architecture also contributes to the reduction
of incurring running costs (Kochut and Beaty, 2007).
CSPs take advantage of these strategies and place
VMs in the cloud architecture such that the least num-
ber of PMs is powered on to reduce energy costs(Yang
et al., 2012).
Among the literature studied, the article (Yang
et al., 2012) is one of the most cited that focuses
on energy consumption minimization to save costs.
The VM placement algorithm proposed in the article
(Yang et al., 2012) considers both workload character-
istics and energy consumption. It is modelled using
modified bin packing problem, where PMs are bins
and VMs are items. The items are further categorized
into red and green items for data-intensive and CPU-
intensive VMs respectively. To save energy, the algo-
rithm consolidates both red and green items into the
same bins of capacity 1. To achieve this, it first places
red items in red bins of capacity C and then the green
items into green bins of capacity (1- C). It then com-
bines the red and green items and places them into
the first bin of capacity 1. If there is a need for mi-
grations, the VM placement algorithm allows migra-
tions of only the green items (CPU-intensive VMs).
This strives to avoid performance degradation which
results from migrating data-intensive VMs away from
the PMs that contain their images. The data-intensive
VMs lose their performance up to 40 percent if they
are migrated away from those PMs that contain their
images.
In summary, the prominent point of the proposed
VM placement algorithm (Yang et al., 2012), based
on the perspective of the paper in hand, is that per-
formance might be compromised if consolidation of
VMs to reduce energy consumption does not consider
placement of VMs based on their workload character-
istics. The VM placement algorithm discussed does
not only save the costs but also strives to maintain the
performance of the cloud environment even after any
migrations. It however, does not take into account
some security risks that are introduced by consolida-
tion of VMs into the same VMs. Some potential se-
curity risks that can be imposed by this kind of VM
placement, amongst others, are the following:
• Consolidation of VMs may result in adversary
users sharing the same PM. This brings higher
risks of inter-VM attacks such as Denial of Ser-
vice (DOS) attacks. The malicious user who owns
one of the VMs consolidated within a PM can
willingly take all resources of the host and deny
access to other co-located VMs (Wu et al., 2010).
• Although virtualization provides isolation be-
tween VMs, there are vulnerabilities exposed by
narrow interfaces in the hypervisor which might
result in possible fault leakages between VMs
consolidated within the same PM (Soltesz et al.,
2007).
2.4 VM Placement Algorithm Focusing
on Security
Security in the cloud environment is a major concern,
more especially in IaaS cloud architecture. This is
due to the fact that IaaS supports multi-tenant services
where users share physical resources leading to po-
tential inter-VM attacks (Caron et al., 2013). There
are some on-going research efforts which strive to
minimize these cloud specific attacks using strategic
placement of VMs.
In the article (Afoulki et al., 2011), authors pro-
pose one of the first and most cited VM placement al-
gorithm that considers security as an objective. This
security-aware VM placement algorithm makes use
of adversary lists submitted by clients to create in-
compatibility groups. The idea is to place VMs in
such a way that no adversary users share the same
physical resources. Upon request for a new VM, the
VM placement algorithm checks the available PMs
for VMs belonging to adversary users. If such VMs
are available, the algorithm checks if it is possible to
either suspend or migrate those VMs. The suspen-
sion of running VMs depends on the priorities of both
the running and the newly requested VM. If both sus-
pensions and migrations are not possible, the algo-
rithm checks if the new VM can be scheduled for later
placement, otherwise the VM request is rejected. In
cases where either preemptions or migrations are pos-
sible, the placement algorithm further checks if the
new VM requirements fit the resource capacities of
the available PM. If the requirements fit resource ca-
pacities, the VM request is accepted and scheduled
to run immediately. In summary, the salient points
of the discussed VM placement algorithm (Afoulki
et al., 2011), based on the perspective of the paper
in hand, are: 1) the consideration of adversary users
and creating of incompatibility groups in the place-
ment of VMs and, 2) the suspension and/or migration
of running VMs, which are based on the priorities as
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
414

compared to new VM request, if there is there is no
available PM that suits the requirements of the new
VM request.
Some other closely related papers (Al-Haj et al.,
2013), (Caron et al., 2013) propose VM placement
algorithms that allow users to choose the security lev-
els they require. The VM placement algorithms in
this case take user requirements as inputs and pro-
vide VM placement accordingly. Yuchi and Shetty
(Yuchi and Shetty, 2015) also propose a VM place-
ment algorithm that separates VMs according to their
vulnerability status. These research efforts minimize
the malicious attacks that are specific to the cloud en-
vironment using strategic placement of VMs. The
discussed research efforts however, focus solely on
security-aware placement of VMs and ignore the QoS
and cost optimization objectives. An important se-
curity feature that can further be augmented to im-
prove the security-aware VM placement algorithms
(Afoulki et al., 2011), (Yuchi and Shetty, 2015), (Al-
Haj et al., 2013), (Caron et al., 2013) discussed in this
subsection is:
• The use of databases (National Vulnerability
Database) of known vulnerabilities to test surviv-
ability status of the PMs. In this case, the sur-
vivability status of the PMs will periodically be
checked instead of checking only when there is
a request for new VM as proposed in (Yuchi and
Shetty, 2015). If the survivability drops for any
PM, there should be an alert that triggers the mi-
gration of critical VMs to more secure PMs. This
will ensure the availability of the critical VMs
which if compromised, can cause massive harm
to the users.
In summary, this paper categorizes the VM place-
ment algorithms into three: VM placement algorithms
that focus on QoS; VM placement algorithm that fo-
cus on cost; and VM placement algorithms that fo-
cus on security. The first two categories ensure op-
timal service delivery at reduced costs. Their strate-
gic placement however imposes some security risks
which can lead to cloud specific attacks. It is there-
fore necessary to enhance the security features to min-
imize these risks. The third category considers secu-
rity as its major objectives, but with possibilities to be
further improved. The next sections discuss and eval-
uate VM placement algorithms to find the one that
qualifies to further be enhanced with security features
or improved from a security point of view.
3 VM PLACEMENT
ALGORITHMS EVALUATION
This section introduces the VM placement algorithms
that are evaluated in this paper. The algorithms and
their brief descriptions are presented in tabular for-
mat as shown in Table 3 in the Appendix. The next
subsections will present the evalution criteria and the
obtained results after evaluation.
3.1 Evaluation Criteria
The idea is to evaluate the currently implemented
VM placement algorithms to identify the algorithm
that qualifies to be used in implementation of O-Sec
VM Placement algorithm. This is achieved by study-
ing the currently developed VM placement algorithms
and evaluating the reflected optimization objectives
for every algorithm. The overall evaluation of the VM
placement algorithm depends on the scaling of the ob-
jectives for every VM placement algorithm.
3.2 Selection of VM Placement
Algorithms
To achieve the goal of this paper, a number of publica-
tions that focus on VM placement algorithms are cho-
sen among the many that are currently implemented.
These publications are found in different websites and
online databases through the use of the internet. Only
those articles published by ACM, IEEE, Springer, El-
sevier, Tailor and Francis, and Science Direct were
considered. The number of publications is reduced by
eliminating those that are not relevant and/or of no in-
terest to the authors of this article. These are selected
through abstract reading of all publications. A thor-
ough studying of the remaining publications narrows
the final selection to only ten publications. These are
the publications which focus on VM placement algo-
rithms evaluated in this paper.
3.3 Step-by-step Evaluation of the
Algorithms
1. Identify the optimization objectives required for
the O-Sec VM placement algorithm.
2. Rate the objectives according to their importance,
using constant values 1 to 4 which represent the im-
portance of the objective.
4 = critical importance,
3 = very important,
2 = important, and
1 = necessary
Towards Optimized Security-aware (O-Sec) VM Placement Algorithms
415

3. Identify the VM placement algorithms, state their
merits, and then evaluate the objectives for every al-
gorithm using the scale below:
2 = objective exists fully,
1 = objective partially exist, and
0 = does not exist
4. Use the formula to score each VM placement algo-
rithm to find those that qualify to be modified in the
implementation of O-Sec VM placement algorithm.
3.4 VM Placement Algorithms
Evaluation
Step 1: Identifying Optimization Objectives for
O-Sec VM Placement Algorithm
For the VM placement algorithm to qualify for the
final selection, certain optimization and security ob-
jectives need to be taken into consideration. The
identification of these objectives depends on their im-
portance, taken from the previous studies (Suseela,
2013), (Vu and Hwang, 2014), (Andreolini et al.,
2009), (Shneiderman, 1984), (Biran et al., 2012),
(Caron et al., 2013). Next are the optimization and
security objectives and motivation on why they are
chosen among the rest.
• Elasticity: PMs within the cloud architec-
ture have limited amount of physical resources
(Suseela, 2013). With the increasing workloads
within the VMs, which are hosted in PMs, there
are chances of resource demands exceeding the
limit. In such cases, there is a need to migrate
some VMs with higher demands to other PMs that
can accommodate the workload (Andreolini et al.,
2009). It is therefore critically important to take
migrations into consideration to allow changes in
workloads.
• Energy Consumption: The rapid increase of en-
ergy consumption in cloud architectures is influ-
enced by the increasing demand for cloud com-
puting resources. Energy consumption is a con-
tributing factor to cost in the cloud, this means
high energy consumption results in high manage-
ment costs. According to (Vu and Hwang, 2014),
energy consumption in clouds is a worldwide con-
cern. There are therefore some on-going research
efforts that concentrate on trying to find different
ways to save energy. It is therefore extremely im-
portant to include it in this study too.
• Response Time: In order to provide good QoS,
time to complete job is an essential tool. The
unexpected delays in computing frustrates users
(Shneiderman, 1984). In the same manner for
cloud computing services we require services that
take the least time to complete. The response time
in this regard is the time taken by an algorithm to
place the VM into its host. Although it is not a
critical objective, but it is necessary in cloud com-
puting.
• VM Communication: Modern data-centers ex-
perience some dynamic traffic change due to the
adoption of dynamic multi-path routing schemas
and varying peak times (Biran et al., 2012). This
means excessive change in traffic can result in
traffic bursts. It is therefore important to con-
sider the communication between VMs and ways
to minimize these traffic bursts.
• Reduced SLA Violations: Normally in cloud
computing, the clients submits the computing re-
quirements known as SLA (Caron et al., 2013).
For every provisioning of the cloud resources,
these SLAs need to be taken into consideration
not to be violated. It is therefore equally impor-
tant to consider it in the new implementation as it
also helps provision some cloud resources.
• Security-aware Optimization: In addition to
these optimization objectives is the security-aware
VM placement, which is the main objective of the
study. The importance of this objective is to mini-
mize some security risks within the cloud environ-
ment. This includes, but is not limited to, cloud
specific attacks such as inter-VM attacks.
There are some other optimization objectives that are
not considered in this paper even though they are also
important, these are: resource utilization, overheads
minimization, heterogeneity of data-centres, and clus-
tering of VMs.
Step 2: Rating of the Optimization Objectives for
O-Sec VM Placement Algorithm
The rating of the optimization objectives based on
their importance are presented in tabular format
shown in Table 1.
Step 3: Evaluation of the Optimization Objectives
of each VM Placement Algorithm
The next step after rating the required objectives
for the O-Sec VM placement algorithm is to eval-
uate the objectives considered in the implementa-
tion of the currently available VM placement algo-
rithms. The currently implemented VM placement al-
gorithms, their merits and the optimization objectives
evaluation are presented in Table 4 in the Appendix.
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
416

Table 1: Objectives Rating Table.
Objectives Ratings
Elasticity 4
Response time 2
Power Consumption 4
Reduced SLA Violations 3
VM Communication 2
Security 2
Step 4: Scoring VM Placement Algorithms for
Final Selection
The final selection process is based on a process in-
troduced in (Eloff, 1983). In this process, a formula is
used to calculate the percentage score in order to find
the VM placement algorithm that best fit the evalua-
tion criteria. The formula is simplified before usage
to suit the research efforts of the paper in hand.
3.4.1 Deriving the Formula
In order to calculate the required optimization objec-
tives percentage, the equation below can be used:
O
j
=
ObjectiveEvaluation
TotalObjectiveRating
× 100 (1)
O
j
=
∑
N
i=1
x
i
k
i
∑
N
i=1
x
i(TOT)
k
i(TOT)
× 100 (2)
Where, x
i
is the evaluation factor allocated to the pres-
ence of the objective, k
i
is the importance rating of
each objective, x
i(TOT)
is the evaluation factor allo-
cated to the presence of the objective in an ideal sit-
uation, k
i(TOT)
is the importance rating for an ideal
situation, and N is total number of objectives used for
evaluation.
For an ideal VM placement x
i(TOT)
= 2 for all
placement algorithms. This makes,
O
j
=
∑
N
i=1
x
i
k
i
2
∑
N
i=1
k
i(TOT)
× 100 (3)
And therefore,
O
j
= 50
∑
N
i=1
x
i
k
i
∑
N
i=1
k
i(TOT)
(4)
In this paper, k
i(TOT)
is a constant as the importance
of the objectives is known (objectives already rated).
The value for k
i(TOT)
= 18, and N = 6, therefore,
O
j
= 0.46
N
∑
i=1
x
i
k
i
(5)
For all,
1 ≤ j ≤ M
Where M is total number of evaluated placement al-
gorithms
3.4.2 Using the Derived Formula
Using the derived formula, we calculate the objec-
tives percentage for each of the studied VM place-
ment algorithms. The algorithm that has the highest
percentage is assumed to be the one that considers
the most important objectives according to our scal-
ing. After the calculations, it is found that the best
placement algorithm to be modified in order to imple-
ment the O-Sec placement algorithms is the Traffic
and Power-aware VM Placement (TPVMP). The
percentage score of all the algorithms evaluated are
shown in Table 2.
Table 2: VM Placement Algorithms Scores.
VM Placement Algorithm Score
RFFA 7.36
VM Placement Algorithm to
Minimize PM Count
9.2
Dynamic Load Management 3.68
MCRVMP 3.68
SEP-Pack and Dynamic Algo-
rithm
8.28
TPVMP 11.04
Migration Based VM Placement
and Direct Placement
5.52
A security-aware Scheduler 6.9
Modified k-means Clustering in
VM Placement
5.52
Security Metrics Security-aware
VM Placement Algorithm
3.68
Towards Optimized Security-aware (O-Sec) VM Placement Algorithms
417

4 TPVMP AS A CANDIDATE TO
BE CONSIDERED FOR O-SEC
VM PLACEMENT
The results obtained from the evaluation criteria in
this paper show that TPVMP qualifies to be a VM
placement algorithm with potential to be further aug-
mented with security features. The implementation of
TPVMP discussed in article (Vu and Hwang, 2014)
focuses on the following objectives: VM communi-
cation , energy consumption, and elasticity. Taking
into consideration the VM communication objective,
the algorithm strives to reduce the total traffic costs
within the cloud networks by placing communicat-
ing VMs into clusters. These clusters are then placed
in those PMs that are close by to each other. It is
envisaged that this VM communication objective of
TPVMP algorithm can be expanded and employed to
possibly include the following security objectives:
• Initial placement of VMs using TPVMP on virtual
PMs, and then rearrangement of these VMs based
on user security requirements and the co-located
VMs. This must not violate the maximum dis-
tance between communicating VMs considered in
original TPVMP. This requires submission of user
security requirements which state the security lev-
els required using metrics as proposed in (Caron
et al., 2013). In addition to this, users need also
to submit adversary user lists in order to create
incompatibility groups. These groups will help
minimize planned security attacks brought by ad-
versary users owning the VMs located in the same
PM.
• The use of a virtual switch within the hypervisors
which allows creation of vNets can help isolate
user traffic. This concept is discussed by Alfoulki
et al. (Afoulki et al., 2011) and ensures that the
traffic generated by cloud users cannot be seen by
other unauthorized cloud users. This reduces the
probability of interception and therefore strives to
maintain the confidentiality, integrity and some-
times availability of user data that is in transit.
The other objective of focus in TPVMP is energy
consumption. The algorithm addresses this objective
by consolidating VM into high capacity servers and
switching of as many low capacity servers as possi-
ble. This objective goes hand-in-hand with the third
objective of focus in the implementation of TPVMP.
The third objective, Elasticity and/or migrations, con-
sidered in this VM placement algorithm build towards
the consolidation of VMs to save energy. For future
implementation, these objectives conceive as a pos-
sibility that they can also be extended to include the
following security objectives:
• Consolidate the VMs within the high capacity
PMs taking into consideration the security level
requirements of the users, incompatibility lists,
and the maximum allowed distance between the
communicating VMs.
• For a new VM request, there are possibilities of
having PMs that meet user security requirements
containing adversary users. In such cases, migra-
tions and/or suspension of already running VMs
should be taken into consideration. The suspen-
sion of the VMs, on one hand, should depend on
the priority of the new VM compared to the al-
ready running VMs (Afoulki et al., 2011). The
migrations, on the other hand, should not violate
the maximum length constraint of the communi-
cating VMs considered in original TPVMP.
There are other possible security features that can fur-
ther extend the TPVMP but not specific to the ob-
jectives of focus in TPVMP. The possible novice im-
provement to TPVMP that can enhance security to the
placement of VMs is as follows:
The VM placement algorithm will first categorize
VMs into two: critical and non-critical VMs. The ex-
amples of critical VMs are back-end databases that
store sensitive data, and those of non-critical VM are
application engines. The critical VMs will be con-
solidated into the same PMs with a constraint that:
no two communicating critical VMs share the same
PM. The other non-critical VMs will be consolidated
on the other remaining PMs. The placement of both
critical and non-critical VMs, however, needs to en-
sure that the distance constraint between communi-
cating VMs considered in original TPVMP is not vi-
olated. Again the VM placement needs to ensure that
the least number of PMs is used in order to mini-
mize energy consumption. To ensure the availabil-
ity of the critical VMs, PMs containing the critical
VMs will continuously be monitored. An attempt to
initiate communication between any of these critical
VMs in the same PM will be marked as a security
threat. The initiating VM will therefore be migrated
to a different PM to disallow the possible unautho-
rized communication. This proposed security-aware
placement strives to avoid inter-VM attacks through
shared network channels within the hypervisor. It
further ensures the availability of the critical VMs
through strategic placement.
5 CONCLUSION
In this paper, authors have developed an evaluation
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
418

method to compare the optimization objectives re-
flected in the different VM placement algorithms.
This evaluation method uses an equation which scores
the different optimization objectives based on their
importance rating. According to the evaluation
method, the VM placement algorithm which scores
the highest is assumed to have the most important op-
timization objectives. The idea is to further expand
such an algorithm to include security-aware objec-
tives in order to achieve an Optimized Security-aware
(O-Sec) VM placement algorithm.
To achieve their goal, the authors of this pa-
per have categorized VM placement algorithms into
three. These categories are based on what the algo-
rithms aim to accomplish; cost reduction, good QoS
and security. The authors further discuss how the VM
placement algorithms for each category can further be
adapted or modified to accomplish the notion of O-
Sec VM placement algorithm. The optimization ob-
jectives reflected in these VM placement algorithms
help to identify and rate the objectives suitable for
the evaluation criteria. After the evaluation process,
the Traffic and Power-aware VM Placement algorithm
(TPVMP) is found as a potential candidate to further
be augmented with security features.
For future envision, the authors propose the pos-
sible extension to the TPVMP to include some secu-
rity features. This extension takes into consideration
the optimization objectives of TPVMP, which are; en-
ergy consumption, VM communication and elastic-
ity. In addition to these, the proposed implementation
will strive to place critical and non-critical VMs sep-
arately with the provision to detect potential attacks
to the VMs. Whenever there is an alert for potential
attacks, migrations will be used to neutralize the sit-
uation. The proposed VM placement algorithm aims
to ensure the availability of the critical VMs through
strategic placement.
ACKNOWLEDGEMENTS
We would like to thank Moseme Anna Thulo, who is
a sister to the first author, for helping in editing the
article in hand.
REFERENCES
Afoulki, Z., Bousquet, A., and Rouzaud-Cornabas, J.
(2011). A security-aware scheduler for virtual ma-
chines on IaaS clouds. Report 2011.
Al-Haj, S., Al-Shaer, E., and Ramasamy, H. V. (2013).
Security-aware resource allocation in clouds. In
Services Computing (SCC), 2013 IEEE International
Conference on, pages 400–407. IEEE.
Alani, M. M. (2014). Securing the cloud: Threats, attacks
and mitigation techniques. Journal of Advanced Com-
puter Science & Technology, 3(2):202.
Andreolini, M., Casolari, S., Colajanni, M., and Messori,
M. (2009). Dynamic load management of virtual ma-
chines in cloud architectures. In International Confer-
ence on Cloud Computing, pages 201–214. Springer.
Biran, O., Corradi, A., Fanelli, M., Foschini, L., Nus, A.,
Raz, D., and Silvera, E. (2012). A stable network-
aware VM placement for cloud systems. In Proceed-
ings of the 2012 12th IEEE/ACM International Sym-
posium on Cluster, Cloud and Grid Computing (ccgrid
2012), pages 498–506. IEEE Computer Society.
Bobroff, N., Kochut, A., and Beaty, K. (2007). Dynamic
placement of virtual machines for managing SLA vio-
lations. In 2007 10th IFIP/IEEE International Sympo-
sium on Integrated Network Management, pages 119–
128. IEEE.
Caron, E., Le, A. D., Lefray, A., and Toinard, C. (2013).
Definition of security metrics for the cloud comput-
ing and security-aware virtual machine placement al-
gorithms. In Cyber-Enabled Distributed Computing
and Knowledge Discovery (CyberC), 2013 Interna-
tional Conference on, pages 125–131. IEEE.
Chowdhury, M. R., Mahmud, M. R., and Rahman, R. M.
(2015). Implementation and performance analysis of
various vm placement strategies in cloudsim. Journal
of Cloud Computing, 4(1):1.
Dong, Z., Liu, N., and Rojas-Cessa, R. (2015). Greedy
scheduling of tasks with time constraints for energy-
efficient cloud-computing data centers. Journal of
Cloud Computing, 4(1):1.
Eloff, J. H. (1983). Selection process for security packages.
Computers & Security, 2(3):256–260.
Gao, J. and Tang, G. (2013). Virtual machine placement
strategy research. In Cyber-Enabled Distributed Com-
puting and Knowledge Discovery (CyberC), 2013 In-
ternational Conference on, pages 294–297. IEEE.
Ilkhechi, A. R., Korpeoglu, I., and Ulusoy,
¨
O. (2015).
Network-aware virtual machine placement in cloud
data centers with multiple traffic-intensive compo-
nents. Computer Networks, 91:508–527.
Kochut, A. and Beaty, K. (2007). On strategies for dy-
namic resource management in virtualized server en-
vironments. In 2007 15th International Symposium on
Modeling, Analysis, and Simulation of Computer and
Telecommunication Systems, pages 193–200. IEEE.
Kuo, C.-F., Yeh, T.-H., Lu, Y.-F., and Chang, B.-R. (2015).
Efficient allocation algorithm for virtual machines in
cloud computing systems. In Proceedings of the ASE
BigData & SocialInformatics 2015, page 48. ACM.
Li, K., Zheng, H., Wu, J., and Du, X. (2015). Virtual ma-
chine placement in cloud systems through migration
process. International Journal of Parallel, Emergent
and Distributed Systems, 30(5):393–410.
Li, M., Zhang, Y., Bai, K., Zang, W., Yu, M., and He, X.
(2012). Improving cloud survivability through depen-
Towards Optimized Security-aware (O-Sec) VM Placement Algorithms
419

dency based virtual machine placement. In SECRYPT,
pages 321–326.
Lin, C.-C., Liu, P., and Wu, J.-J. (2011). Energy-efficient
virtual machine provision algorithms for cloud sys-
tems. In Utility and Cloud Computing (UCC), 2011
Fourth IEEE International Conference on, pages 81–
88. IEEE.
Lindemann, J. (2015). Towards abuse detection and pre-
vention in IaaS cloud computing. In Availability, Re-
liability and Security (ARES), 2015 10th International
Conference on, pages 211–217. IEEE.
Meng, X., Pappas, V., and Zhang, L. (2010). Improving the
scalability of data center networks with traffic-aware
virtual machine placement. In INFOCOM, 2010 Pro-
ceedings IEEE, pages 1–9. IEEE.
Ohta, S. (2013). Virtual machine placement algorithms to
minimize physical machine count. In APNOMS, pages
1–3.
Shneiderman, B. (1984). Response time and display rate in
human performance with computers. ACM Computing
Surveys (CSUR), 16(3):265–285.
Soltesz, S., P
¨
otzl, H., Fiuczynski, M. E., Bavier, A., and
Peterson, L. (2007). Container-based operating sys-
tem virtualization: a scalable, high-performance al-
ternative to hypervisors. In ACM SIGOPS Operating
Systems Review, volume 41, pages 275–287. ACM.
Suseela, B. B. J. (2013). Survey on VM placement algo-
rithms. International Journal of Engineering Trends
and Technology (IJETT), 6(7):349–352.
Venkata, S., Kumar, K., and Padmapriya, S. (2004). A sur-
vey on cloud computing security threats and vulnera-
bilities.
Vu, H. T. and Hwang, S. (2014). A traffic and power-aware
algorithm for virtual machine placement in cloud data
center. International Journal of Grid & Distributed
Computing, 7(1):350–355.
Wu, H., Ding, Y., Winer, C., and Yao, L. (2010). Net-
work security for virtual machine in cloud computing.
In Computer Sciences and Convergence Information
Technology (ICCIT), 2010 5th International Confer-
ence on, pages 18–21. IEEE.
Xu, J. and Fortes, J. (2011). A multi-objective approach to
virtual machine management in datacenters. In Pro-
ceedings of the 8th ACM international conference on
Autonomic computing, pages 225–234. ACM.
Yang, J.-S., Liu, P., and Wu, J.-J. (2012). Workload
characteristics-aware virtual machine consolidation
algorithms. In Cloud Computing Technology and Sci-
ence (CloudCom), 2012 IEEE 4th International Con-
ference on, pages 42–49. IEEE.
Yuchi, X. and Shetty, S. (2015). Enabling security-aware
virtual machine placement in IaaS clouds. In Mil-
itary Communications Conference, MILCOM 2015-
2015 IEEE, pages 1554–1559. IEEE.
Zaouch, A. and Benabbou, F. (2015). Load balancing for
improved quality of service in the cloud. International
Journal of Advanced Computer Science and Applica-
tions IJACSA, 6(7):184–189.
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
420

APPENDIX
Table 3: VM Placement Algorithms.
VM Place-
ment Algo-
rithm
Brief Summary
Resource-
based
First-Fit
Algorithm
(RFFA)
These algorithms, proposed in (Kuo et al., 2015) allocate VMs according to the re-
source requirements and the availability of resource amounts in the running PMs. To
save energy, the algorithms minimize the number of running PMs by switching off
those that are on idle mode. The algorithms assign new VMs to the first PMs that best
suit the requirements. There are some possible migrations if the resource requirement
of the running VMs change and exceed the resource capacity of the hosting PM
VM Place-
ment Al-
gorithm to
Minimize
PM Count
The proposed heuristic VM placement algorithm by (Ohta, 2013) rearranges VMs
within the cloud environment in response to the load changes. The algorithm is made
up of three procedures: the judgestate, incrstate and decrstate. These procedures first
determines the state of the PMs within the cloud architecture, then choose the sending
and receiving PMs for migrations to take place in order to increase or decrease number
of PMs respectively
Dynamic
Load Man-
agement
The proposed VM placement algorithm (Andreolini et al., 2009) strives to reduce over-
heads caused by excessive migrations. To achieve this, the algorithm considers the load
profile of the PM and load trend behaviour of the VMs over time to avoid false alarms
that trigger migrations. The migrations are triggered only by changes that are signif-
icant and persistent. For migrations to take place, the algorithm identifies those hosts
that are over-utilized then sorts VMs in those hosts according to their loads. Then the
subset of VMs with highest loads is chosen for migration.
Min Cut
Ratio-
aware VM
Placement
(MCRVMP)
The proposed VM placement algorithm (Biran et al., 2012) addresses issues of unpre-
dicted traffic bursts. This solution achieves its goal by placing VMs in such a way that
there is spare capacity in every network cut. This solution consists of two algorithms
that cluster VMs into CCs. They place VMs with high communicating weight in PMs
that are in close proximity.
SEP-Pack
and Dy-
namic
Algorithm
The proposed VM placement algorithm (Yang et al., 2012) considers both energy con-
sumption minimization and performance. It is a modified bin-packing algorithm where
VMs are items and PMs are bins. The items are further categorized into two: red items
for data-intensive VMs and green items for CPU-intensive VMs.
Traffic
and Power
Aware VM
Placement
(TPVMP)
The proposed VM placement algorithm (Vu and Hwang, 2014) considers both traffic
and power consumption. In order to save energy, the algorithm consolidates VMs into
high capacity servers and tries to switch off as many low capacity servers as possible.
To reduce traffic costs within the cloud architecture, the algorithm tries to place VMs
with high traffic weight in PMs that are close range to each other.
Migration
Based VM
Placement
The proposed VM placement algorithm (Li et al., 2015) considers total completion
time to place a VM. The placement algorithm has two options: migration based place-
ment, and direct placement. For migration based placement, VM is placed in a PM
that takes least job completion time, but if there are limited resources in that PM to
host the new VM, then some already hosted VMs within the PM are migrated.
A security-
aware
Scheduler
The proposed VM placement algorithm (Afoulki et al., 2011) takes lists of adversary
users from clients as input. These lists are used to create incompatibility groups used
by scheduler to place VMs. The algorithm ensures that no adversary users reside in
the same PM.
Modified k-
means Clus-
tering in VM
Placement
The proposed VM placement algorithm (Chowdhury et al., 2015) explores the cluster-
ing technique where VMs are clustered according to their CPU utilization and RAM
allocation. The algorithms migrates VMs from over-utilized or under-utilized PMs
using the modified k-means centroid clustering. Unlike the normal k-means centroid
clustering where the initial centroid is random, this algorithm makes the initial cen-
troid the average of the n-objects, which are VMs in this case. After clustering the
VMs, the bin-packing algorithms are used to allocate VMs, starting with those that are
in the highest dense cluster.
Security-
aware VM
Placement
Algorithm
The VM placement algorithm (Caron et al., 2013) uses metric vectors to allow users
to detail their security requirements, presented as bitsets. The clients submit their
security requirements using the bitsets which are easily interpreted by the scheduler.
Towards Optimized Security-aware (O-Sec) VM Placement Algorithms
421

Table 4: Objectives Evaluation Table.
VM Placement Algorithm Merits Objective Evaluation
Resource-based First-Fit
Algorithm (RFFA)
The VMs are placed according to their
resource requirements. The idle PMs
are switched off to save energy
Elasticity 2
Response Time 0
Power Consumption 2
SLA Violations 0
Communication 0
Security 0
VM placement algorithms
to minimize PM count
Live migrations allow rearrangement of
VMs in response to load changes. This
helps minimize number of running PMs
Elasticity 2
Response Time 2
Power Consumption 2
SLA Violations 0
Communication 0
Security 0
A Security-aware
Scheduler
The adversary users do not share the
physical machine. vNets are created to
isolate traffic
Elasticity 2
Response Time 0
Power Consumption 0
SLA Violations 2
Communication 0
Security 2
Security metrics
security-aware VM
placement algorithm
Ability to test the security levels of the
physical machines, allows users to state
the security levels they require using
metrics
Elasticity 0
Response Time 0
Power Consumption 0
SLA Violations 2
Communication 0
Security 2
Dynamic Load
Management
The algorithm ignores the absolute or
average load state, but triggers
migrations depending on the load
profile over time
Elasticity 2
Response Time 0
Power Consumption 0
SLA Violations 0
Communication 0
Security 0
Min Cut Ratio-aware VM
Placement (MCRVMP)
Addresses issues of unpredicted traffic
bursts
Elasticity 0
Response Time 0
Power Consumption 0
SLA Violations 0
Communication 2
Security 0
SEP-Pack and Dynamic
Algorithm
Optimizes the use of resources by
ensuring placement both data-intensive
and CPU-intensive VMs in every PM
Elasticity 2
Response Time 1
Power Consumption 2
SLA Violations 0
Communication 0
Security 0
Traffic and Power Aware
VM Placement (TPVMP)
Tries to reduce the traffic costs within
the datacenter while preserving energy
Elasticity 2
Response Time 0
Power Consumption 2
SLA Violations 0
Communication 2
Security 0
Migration Based VM
Placement and Direct
placement
Tries to minimize the total completion
time to place the new VM
Elasticity 2
Response Time 2
Power Consumption 0
SLA Violations 0
Communication 0
Security 0
Modified k-means
Clustering in VM
Placement
Concentrates on placement of VMs that
are in the migration list. Modifies the
available bin-packing solution
algorithms
Elasticity 2
Response Time 0
Power Consumption 1
SLA Violations 0
Communication 0
Security 0
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
422