Journal of Ambient Intelligence and Humanized Computing

https://doi.org/10.1007/s12652-022-04439-y

ORIGINAL RESEARCH

OSVR: an efficient support vector regression model based host

overload detection and secure virtual machine migration
S. Parthasarathy1

Received: 15 May 2021 / Accepted: 28 September 2022

© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2022

Abstract
The utilization of cloud computing server farms is developing quickly, and the interest in high-performance computing
(HPC), stockpiling, and systems administration assets for business and logical applications is expanding. Virtual machine
(VM) coordination includes the immediate migration of VMs running on fewer physical servers, accordingly permitting
more workers to be killed or running in low force mode to improve power utilization proficiency, working expense, and
CO2 emanations. A significant advance in VM reconciliation is the location of the host over-burden, which endeavors to
anticipate whether the physics server will over-subscribe with VMs. For overloaded detection, in this paper optimized sup-
port vector regression (OSVR) model is proposed. Here, to enhance the performance of support vector regression (SVR),
the parameters of SVR are optimally selected by using the Oppositional Beetle Swarm Optimization (OBSO) algorithm.
After the overloaded detection, the virtual machines are migrated. However, during migration, the system integrity and
confidentiality of data may be affected. To avoid the problem, before migration, the data are encrypted using Cyclic Shift
Transposition Algorithm (CSTA). The performance of the proposed approach is analyzed in terms of different measures
and effectiveness compared with different methods. The results shows that our proposed approach attained the minimum
migration and energy consumption.

Keywords Overloaded · Migration · Cloud computing · Consolidation · Encryption · Security · OSVR · EBSO · Cyclic
Shift Transposition Algorithm

1 Introduction (Beloglazov 2013; Leelipushpam and Sharmila 2013; Medina

and García 2014). One of the principal factors in energy waste
Cloud computing, also called Internet-based computing and is that too many idle PMs exist. Thus, ensuring that the lowest
provides on-demand computing resources and data to com- possible number of PMs is active is an efficient approach to
puters and any web-connected devices. Using various tech- reducing energy expenses in cloud data centers (Monil and
nologies and concepts (e.g., hardware virtualization and data Rahman 2015). Recently, several studies and developments
centers), cloud computing can achieve economies of scale have been proposed for decreasing the energy expenses of
and is grouped into three cloud service models and four cloud cloud data centers.VM consolidation is commonly split into
deployment models (Beloglazov and Buyya 2013; Thakur the following steps (Melhem et al. 2017): (1) detecting over-
et al. 2013), each of which addresses a different type of busi- loaded hosts; (2) detecting under-loaded hosts; (3) selecting
ness information technology (IT) requirement. Generally, the VMs; and (4) placing VMs. Nowadays, a lot of overloaded
average central processing unit (CPU) utilization of physi- prediction methods are developed. Among that, most of the
cal machines (PMs) is only 15–20% in their common state. research on CPU utilization is considered for the VM migra-
Additionally, idle PMs comprise the majority of PMs and tion process (Mandal and Khilar 2013). Their model proposes
continuously consume 70% of their peak energy consumption that the power consumption of a server grows in line with the
growth of CPU usage. CPU usage-based models can provide
* S. Parthasarathy an accurate prediction for CPU-intensive applications; how-
[email protected] ever, they may not be accurate for other types of applications
such as network, I / O, and memory-intensive applications
1
SRM Valliammai Engineering College, (Beloglazov and Buyya 2010).
Kanchipuram, Chengalpattu, Tamil Nadu 603203, India

13
Vol.:(0123456789)
 S. Parthasarathy

After overloaded detection, the overloaded VMs are consolidation in cloud computing using a probabilistic model.
migrated. VM migration or portability is the critical ele- In this paper, self-versatile architecture was introduced to
ment of virtualization innovation which is utilized to give distinguish and oversee underloaded and over-burden VMs/
equipment/framework support, responsibility adjusting, and PMs in response to responsibility changes in the server. The
straightforward administration in ordinary server farms just objective of consolidation was to utilize the base number of
as in Cloud foundation (Ahmad et al. 2015; Shetty et al. dynamic VMs and PMs while ensuring the quality of service
2012; Oberheide et al. 2008; Kazim et al. 2013). Notwith- (QoS). Moreover, Rasouli et al. (2020) developed learning
standing, VM migration with metadata (keys of scrambled automata-based energy-efficient virtual machine consolidation
circle mages) isn't secure as a result of the inaccessibility of on the cloud. Here, they introduced an extended-employment
solid security includes in hypervisors.VM migration with- position by learning Automata (EPPLA) as a dynamic alter-
out security becomes a single purpose of a disappointment native to VMs on server farms to reduce power consumption.
for Cloud climate since interlopers can infuse malevolent Reproduction results showed that EPPLA's energy use var-
code or alter the VM content. Effective assault on migra- ied by 20% and 30% and that PPLA and Firefly were shown
tion interaction may cause refusal of administration (DOS), separately. El-Moursy et al. (2019) explained host overload
loss of information respectability, and secrecy in sent VM detection using multi-dimensional regression host utilization.
(Wang et al. 2010; Chen et al. 2012). To maintain a strategic Three factors are considered for many setbacks, namely CPU,
distance from the assault, before migration the information memory, and network BW usage.
ought to be gotten. Chen et al. (2019) had explained the overload detection
The main aim of this paper is to effectively find out the using a typical weighted moving average (GWMA). This
overloaded PM and securely migrate the VM to reduce algorithm is used to detect overloaded hosts, which deals
energy consumption. Here, for host overload detection with the dynamic integration of VMs based on the analy-
OSVR model is utilized. After overload detection, the sis of historical data on the resource usage of VMs. This
migrated VM data are encrypted using CSTA. The perfor- increases the accuracy in calculating the top threshold for
mance of the proposed approach is analyzed based on differ- host overloading, which in turn increases the accuracy in
ent metrics. The main contribution of the proposed approach identification to deal with the migration problem of VMs.
is listed below; Hezavehi et al. (2020) had developed a DTOS attack detec-
tion framework based on an anomaly in the cloud envi-
• Propose an optimized support vector regression model ronment using a third-party auditor (DPA). Li and Zhihua
for host overloaded detection. In this multi-objective (2019) developed the Markov decision process-based over-
function is designed. The multi-objective function is load detection. With the solution of the Pelman optimal
based on three parameters namely, CPU utilization, equation improved by the value repeat method, the optimi-
memory usage, and network bandwidth. zation model was solved and the optimal overload gate was
• To enhance the SVR performance, the parameters of selected with adaptation. Wu et al. (2017) had developed
SVR are optimally selected using oppositional beetle secure aware virtual machine migration. Here, the proxim-
swarm optimization (OBSO). The OBSO is a combina- ity integrity measurement (AIM) dynamically monitors the
tion of oppositional-based learning (OBL) and BSO. proximity of its adjacent hypervisor and sends the results to
• To avoid data loss during the period of migration, the the regional complex TCB, which then determines whether
data are encrypted using CSTA. the migration can begin and where to migrate according to
• The performance is analyzed in terms of different met- the integrity verification table.
rics. The above literature surveys are mainly focused on migra-
tion and security in the cloud. All the methods have given
The rest of the paper is organized as follows; the literature better results. Even though, the methods have some limita-
survey is presented in Sect. 2 and proposed host overload tions namely, high response time, security leakage, and an
detection and secure migration are presented in Sect. 3. The SLA violation. To avoid the problem, in this paper, secure
experimental results are analyzed in Sect. 4 and the conclu- virtual machine migration is proposed.
sion part is presented in Sect. 5.

3 Proposed host overload detection

2 Literature survey and secure virtual machine migration

Several researchers had developed host overload detection, The high demand for computing power by business, science,
consolidation, and migration. Among them, a few works are and applications has prompted the creation of energy-inten-
analyzed here; Abadi et al. (2018) introduced virtual machine sive data centers that swallow enormous amounts of energy

13
OSVR: an efficient support vector regression model based host overload detection and secure…

activity. This high energy use forces a huge labor cost, as 3.1 SVR‑based host overload detection
well as affects the environment (greenhouse gas emissions).
One promising answer to reducing the amount of energy The host utilization is a unique time series. By analyzing
used by server farms is the integration of virtual machines the time sequence patterns, host overload and underload
(VMs), which allow a few hosts to enter low-swallowing rest predictions for the host are possible because future values
systems. Dynamic migration (replacement) of VMs between over the same time series are based entirely on past obser-
actual hosts is a technique for accomplishing VM integra- vations. The goal of SVM regression is to detect a func-
tion. Replacing VMs hosted by inactive hosts not only saves tion that provides the greatest ɛ deviation from the target
dynamic migration energy but also avoids areas of interest values, so the maximum allowable error is ɛ.To predict
by moving VMs from overused hosts. Before moving VMs, the future values, at first, the time series u1 , u2 , ...un data
it is important to find hosts with high loads. To find the over- are split {(
into training
) inputs and} outputs. Let the training
loaded host, in this paper, optimized SVR is proposed. For set O = ui , = vi , i = 1, ..., N be given, where ui ∈ R
this multi-objective function is designed. After the overload refer the input vector, v ∈ R refer the output vector. It is
detection process, before migration, the data are encrypted. required to find a function f (u) from some specific class
The overall process of the proposed methodology is given K which minimizes the value of empirical risk (predic-
in Fig. 1. tion error on the training set). For the SVM-regression
In Fig. 1, initially, the input tasks are given to the load construction algorithm the function f (u) can be written
balancer. The load balancer is used to allocate the task on in general form as;
resources based on the load of the resources. To avoid the
overload on the host, the OBSO algorithm is presented. F(u) = sT u + c (1)
During the migration, to avoid security issues, the CSTA
where, c represent the Offset of the regression line and s
algorithm is presented. The proposed approach is deeply
represent the Slop.
explained in the below section.

Fig. 1  Overall structure of proposed methodology

13
 S. Parthasarathy

( ) ( )
In this (1), values c and s are calculated. For calculat- ‖ ‖
K ui , uj = exp − ‖ui − uj ‖ ∕ 2𝜎 2 , 𝜎 > 0 (10)
ing c and s the following Eq. (2) should be minimized ‖ ‖

1 C ��
n
� �� The parameters used in this SVR are C, 𝜎 and ∈ to be
R= ‖s‖2 + �vi − f ui � (2) optimized in the training stage. To optimize the parame-
2 n i=1 � �∈
ters, in this paper oppositional beetle swarm optimization
(OBSO) algorithm is utilized.
⎧ � � ��
� � � � ⎪ 0 if � v − f ui � ≤∈
�vi − f ui � = ⎨ � � i �
� �∈ � � � (3)
⎪ �� i
v − f ui �� − ∈ Otherwise 3.2 Parameter optimization using OBSO
⎩
To enhance the performance of the SVR model, the param-
This issue can be formulated in a dual space by
eters are optimally selected. For parameter selection, in this
( ) 1 ∑(
n
)( ) paper OBSO algorithm is utilized. OBSO is a combina-
Maximize LP 𝛽i , 𝛽i∗ = − 𝛽i − 𝛽i∗ 𝛽j − 𝛽j∗ uTi uj tion of oppositional-based learning (OBL) and BSO. The
2 i,j=1
step-by-step process of OBSO-based parameter selection is
∑
n
( ) ( ) explained below;
− ∈ 𝛽i + 𝛽i∗ + 𝛽i − 𝛽i∗ vi
i−1
Step 1: Solution encoding: solution encoding is an essen-
(4) tial step for optimization problems. Initially, the parameters
�n � � of BSO algorithms are initialized. Then the positions of bee-
⎧ 𝛽i − 𝛽i∗ = 0 tles or candidate solutions are initialized with the population
⎪ i=1
Subject to = ⎨ 0 ≤ 𝛽i ≤ C, i = 1, ...., n (5) size N. After the parameter initialization, the initial solution
⎪ ∗ is randomly generated. In this section, C, 𝜎 and ∈ values are
⎩ 0 ≤ 𝛽i ≤ C, i = 1, ....., n
randomly initialized. The solution is defined as P.
where 𝛽i , 𝛽i∗ ≥ 0 represent the Positive Lagrange multipliers, Step 2: Opposite solution generation: Based on the initial
C represents the positive parameter. The best liner hypersur- solution, the opposite solution is generated. Firstly, an initial
face regression is given in Eq. 6. solution is given, and next the corresponding reverse initial
solution is produced. This method will help to enhance the
∑
n
( ) efficiency of BSO and to improve the quality of solutions.
F(u) = sTo u + c = 𝛽i − 𝛽i∗ uTi u + c (6)
i=1
The opposite positions (OPi ) are completely defined by
components Pi.
∑ [ ]
n
( ) OPi = op1i , op2i , .....opdi (11)
s0 = 𝛽i − 𝛽i∗ ui (7)
i=1
w h e [r e OPi =]Loweri + Upperi − Pi with
To create a linear regression (LR) hyperplane, in this OPi ∈ Loweri , Upperi and ­Pi represent the set of solution.
non-linear regression (NLR) model, the input information Step 3: Fitness calculation: After solution initialization,
is converted to high dimensional feature spacing based on the fitness of each solution is calculated. In this paper, a
the kernel function. In the NLR model, the learning prob- multi-objective function-based fitness function is designed.
lem is once again generated in the linear case. The NLR The designed multi-objective function is a combination of
hyperplane function can be written as, CPU utilization, memory, and Bandwidth.
∑
n
( ) ( ) a1 a2 a3
𝛽i − 𝛽i∗ K ui , u + c Host utilization = × ×
F(u) = (8) 1 − CPU 1 − RAM 1 − BW
i=1 (12)
( )
The kernel function K ui , u can be written as follows; where, ai represent the weight factor of CPU, RAM, and
( ) ( ) ( ) BW, CPU represents the relative CPU utilization, RAM rep-
K ui , uj = ΦT ui Φ uj i, j = 1, 2, ...., N (9) resents the Relative memory utilization and BW represents
the Relative Bandwidth utilization.
where,
( ) ( ) Step 4: Updation using BSO: After the fitness calculation,
Φ ui , Φ uj → Feature space of ui and uj. solutions are updated using the OBSO algorithm. The BSO
For kernel calculation, in this paper, a Radial basis ker- algorithm is used to ensure that beetle individuals indepen-
nel (RBF) is utilized. The RBF is defined as follows; dently learned information about their surroundings for local
search. With this learning mode, people are moving towards

13
OSVR: an efficient support vector regression model based host overload detection and secure…

an optimal solution to the problem. The mathematical model authority like the main management system. Using this
for simulating its behavior is as follows; can avoid major related issues. This method consists of
two steps namely, encryption and decryption. The encryp-
Xjst+1 = Xjst + 𝜆Qtjs + (1 − 𝜆)𝜀tjs (13) tion process is used to convert the original data into a
cyber test. And decryption is the process of converting
where; t represents the current number of iterations, Qjs rep- cybertext into original data without losing original infor-
resent the Speed of beetles 𝜀js represent the beetle position mation. The step-by-step process of CSTA is explained
movement. The speed of the beetle is calculated as follows; below;
Qt+1
js
= 𝜔Qtjs + c1 r1 (Ujst − Ptjs ) + c2 r2 (Ugs
t
− Ptgs ) (14)

where, ­c1 and ­c2 represent the two optimistic constants, r­ 1, 3.3.1 Encryption process
and ­r2 represent the arbitrary functions inside the range [0,1],
and ϖ represents the inertia weight. The encryption process is used to hide the original infor-
This paper adopts the strategy of decreasing inertia mation. The proposed CSTA performs the partitioning and
weight, which is mathematically represented as follows; shifting functions, such as row transition, column transition,
primary diagonal transformation, and secondary diagonal
𝜔max − 𝜔min
𝜔 = 𝜔max − ∗t (15) transformation. The encryption process is explained below;
T Step 1:- Consider the input document Ai. To start the pro-
where, cess, the input document Ai is converted into N × N matrix
𝜔min , 𝜔max→ denotes the minimum as well as the maxi- format.
mum value of 𝜔, Step 2:- After that, the shift column (SC) operation is
T, t → denotes the present number of iterations as well as applied to the N × N matrix. The SC calculation is given
the maximum number of iterations, in Eq. (18).
Here, the 𝜉 function defines the incremental function
A�r,c = Ar + shift(r,Mb ) Mb,c (18)
which is mathematically represented as follows; mod
( )
𝜉jst+1 = 𝛿 t ∗ Qtjs ∗ sign(f (Ptrs ) − f (Ptls )) where shift r, Mb depends only on the key value. The key
(16)
values vary from 0 to 9. Key values denote the number of
In this step, high dimension solutions are updated. Here elements that need to be shifted and the mod represents the
𝛿 indicates the step size. The search behaviors of the right arithmetic function.
antenna and the left antenna are respectively expressed as: Step 3:- After the SC operation, shift row (SR) operation
is applied. The SR calculation is given in Eq. (19).
Xrst+1 = Xrst + Qtjs ∗ d∕2
(17) A�r,c = Ar,c + shift(r,Mb ) mod Mb , c (19)
Xlst+1 = Xlst − Qtjs ∗ d∕2
Step 4:- Then, the diagonal shift (DS) operation is per-
Step 5: Termination: The above operations are continued formed. In DS operation, the diagonal elements are shifted
until finding the optimal solution or optimal SVR param- from top to right bottom. The DS function can be written
eters. Once the optimized parameters are obtained, then the as follows;
algorithm will be terminated. This optimized parameter is
used as the host overload decision-maker which reduces A�r,c = Ar + shift(r,Mb ) mod Mb,c+shift(r,Mb ) mod Mb (20)
delay, data loss, and energy consumption.
Step 5:-Again DS operation in a particular order is per-
3.3 Secure migration using CSTA formed. The function can be written as follows;

A�r,c = A(r−1) mod Mb,c. (21)

After host overload detection, the migrated VM data
are encrypted for security purposes. Because, during Step 6:- Then, find the output in a particular order.
the migration time, the malicious maybe attack the VM
and collapse the sensitive information stored on VM. To A�r,c = A(c+(Mb−1 )),c. (22)
avoid the problem, before the migration process, data
are encrypted. For encryption, in this paper, the CSTA Step 7:- After that, the output is converted into ASCII
approach is utilized. The proposed CSTA algorithm format to get the encrypted text.
increases the confidentiality of the data. The proposed Step 8:- Finally, the hash value with the timestamp is
CSTA algorithm does not depend on any centralized calculated and stored in the cloud.

13
 S. Parthasarathy

3.3.2 Decryption process Throughput: Throughput is measured based on a per-

centage of packets successfully delivered to the receiver
The decryption process is the reverse of the encryption or destination.
process. The step-by-step process of decryption is listed Service utilization: It refers to the use of the service used
below; for a given number of requests.
Step 1: At first, the hash values with the time stamp
from the encrypted data are calculated. 4.2 Experimental results
Step 2: Then, encrypted data are converted into ASCII
format. The main objective of this proposed approach is to identify
Step 3: After that, SR operation is applied in a specific the overloaded and underloaded host to decrease the energy
order. consumption and migrate the overloaded host VM securely
Step 4: Then, SC operation is applied in a specific order. to the underloaded host. The experimental results obtained
Step 5: Then, DS operation is applied diagonally. from the proposed and existing approaches are listed below;
Step 6: After that, again DS operation is applied to the In Fig. 2a, the performance of the proposed approach is
output in a specific order. analyzed in terms of migration SLA violation by varying
Step 7: Finally, the decrypted output is obtained. numbers of VMs. The minimum number of migrants should
From the above process, the input task is encrypted. be a good system. When analyzing Fig. 2a, compared to
After encryption, the data will be migrated. other methods, the proposed approach is taken only mini-
mum number of migrations. Here, as the number of VMs
increases, the migration also gradually increases. In Fig. 2b,
the performance of the proposed approach is analyzed in
4 Result and discussion terms of energy consumption. Here, the number of VMs
increases; the energy consumption also gradually increases.
The proposed overloaded Host detection approach experi- When analyzing Fig. 2b, proposed OBSO + SVR and
mental results are analyzed in this section. The proposed BSO + SVR performance is better than GA + SVR and SVR-
method is implemented using Cloudsim with JAVA. The based host overload detection. Compared to BSO + SVR-
performances are analyzed in terms of different evaluation based overload detection, the approach is slightly better. This
metrics and the effectiveness of the method compared with is due to OBSO-based parameter optimization. OBL strategy
different host detection methodologies. enhances the performance of the individual BSO approach.
In Fig. 3a, the performance of the proposed approach is ana-
lyzed in terms of SLA violation. The good system attains
4.1 Evaluation metrics the minimum violations. When analyzing Fig. 3a, the pro-
posed approach is taken minimum SLA violation compared
The performance of the proposed methodology is analyzed to existing. Similarly, in Fig. 3b, the performance of the
in terms of different metrics namely, total energy con- proposed approach is analyzed in terms of ESV by a varying
sumption, number of VM migrations, and SLA violation. number of VMs. Here, also proposed approach attained bet-
Total energy consumption: It is defined as the total ter results compared to the existing approach. This is due to
energy consumed by the physical resources of a data OBSO-based parameter optimization. From the results, it’s
center as a result of application workloads. clear that the proposed approach attained better performance
The number of VM migrations: After high load and compared to existing approaches.
loaded hosts are selected, VMs are selected for migration. In Fig. 4a, the performance of the proposed approach is
Reduction time is the most important barrier in the migra- analyzed in terms of encryption time by varying data sizes.
tion step, and this is achieved by reducing the total number The time taken for encrypting the original file is called
of VM migrations. encryption time. Here, the file size varies from 16 to 256
SLA violation: It catches performance degradation due bits. Here, the proposed CSTA is compared with different
to performance degradation due to host overloading and crypto algorithms namely, AES and DES. When analyz-
VM migration. SLA Violation (SLAV) occurs when a VM ing Fig. 4a, proposed CSTA-based encryption consumes
fails to meet its Promised Service Quality (QoS). less time compared to AES and DES algorithms. As per
Encryption time: The time taken for encrypting the analysis, encryption time increases as a file size increases.
original file is called encryption time. It has been decided that the CST method uses less com-
Decryption time: The time taken for decrypting cipher- putational time and is more suitable for processing jobs in
text is called decryption time. a distributed environment. In Fig. 4b, the performance of
the proposed approach is analyzed in terms of decryption

13
OSVR: an efficient support vector regression model based host overload detection and secure…

Fig. 2  Performance analysis based on a Migration and b Energy consumption

Fig. 3  Performance analysis based on a SLA and b energy and SLA violation (ESV)

Fig. 4  Time analysis a Encryption time b Decryption time

13
 S. Parthasarathy

Fig. 5  Experimental output a 94 b 94

a Throughput and b Service
utilization 92 92
90 90

Service utilization(%)
Throughput(%)
88 88
86 86
84 84
82 82
80 80
78 78
76 76
74 74
DES AES CSTA DES AES CSTA

time. The time taken for decrypting the ciphertext is called Author contributions The author read and approved the final
decryption time. When analyzing Fig. 4b, the proposed manuscript.
method is taken minimum time to decrypt the data. The Funding The authors declare that they don’t have competing interests
performance of the proposed method is analyzed in terms and funding.
of throughput given in Fig. 5a. When analyzing Fig. 5a,
the proposed method attained the maximum throughput Data availability Data sharing is not applicable to this article as no
datasets were generated or analysed during the current study.
of 93% which is 81% for DES-based security and 85%
for AES-based security. Compared to AES and DES, the Declarations
proposed algorithm achieved the maximum throughput.
Moreover, in Fig. 5b, the performance of the proposed Conflict of interest The corresponding author states that there is no
approach is analyzed in terms of service utilization. Here conflict of interest.
also, the proposed approach obtained better results. From
the results, it’s clear that the proposed approach attained
better results compared to existing methods.
References
Abadi RMB, Rahmani AM, Alizadeh SH (2018) Self-adaptive archi-
tecture for virtual machines consolidation based on probabilistic
5 Conclusion model evaluation of data centers in Cloud computing. Cluster
Comput 21(3):1711–1733
In this paper, the proposed overload detection and secure Ahmad RW, Gani A, Hamid SHA, Shiraz M, Xia F, Madani SA (2015)
Virtual machine migration in cloud data centers: a review, taxon-
migration technique have been developed. To avoid energy omy, and open research issues. J Supercomput 71(7):2473–2515
consumption and CO2 emission, overload detection and Beloglazov A (2013) Energy-efficient management of virtual machines
consolidation are important tasks. For overload detection, in data centers for cloud computing. Ph.D. thesis, The University
the OSVR model has been developed. The proposed OSVR of Melbourne
Beloglazov A, Buyya R (2010) Adaptive threshold-based approach for
model mathematical functions are clearly explained. Here, energy-efficient consolidation of virtual machines in cloud data
to enhance the SVR, the parameter of SVR has been opti- centers. In: Proc. 8th Int. Workshop Middleware Grids, Clouds
mally selected by using the OBSO algorithm. The multi- e-Sci, p 4
objective function is used to achieve the migration process. Beloglazov A, Buyya R (2013) Managing overloaded hosts for dynamic
consolidation of virtual machines in cloud data centers under qual-
The multi-objective function is designed based on three ity of service constraints. IEEE Trans Parallel Distributed Syst
parameters namely, CPU utilization, RAM utilization, and (TPDS) 24(7):1366–1379
Bandwidth utilization. During the migration period, the Chen J-H, Lu S-L (2019) GWMA algorithm for host overloading detec-
data has been encrypted using CSTA. The encryption and tion in cloud computing environment. In: International Confer-
ence on e-Business Engineering, Springer, Cham, pp 358–370
decryption process has been clearly explained. The per- Chen Y, Shen Q, Sun P, Li Y (2012) Reliable migration module in
formance of the proposed approach is analyzed in terms of trusted cloud based on security level design and implementation.
different metrics. In the future, we will more concentrate In: International Parallel and Distributed Processing Symposium
on secure data storage in the cloud. Also, we will con- Workshops & PhD Forum
El-Moursy AA, Abdelsamea A, Kamran R, Saad M (2019) Multi-
centrate Context-Aware VM Placement for Heterogeneous dimensional regression host utilization algorithm (MDRHU)
IaaS Cloud using optimization techniques. for host overload detection in cloud computing. J Cloud Comput
8(1):1–17

13
OSVR: an efficient support vector regression model based host overload detection and secure…

Hezavehi SM, Rahmani R (2020) An anomaly-based framework for Rasouli N, Razavi R, Faragardi HR (2020) EPBLA: energy-efficient
mitigating effects of DDoS attacks using a third party auditor in consolidation of virtual machines using learning automata in
cloud computing environments. Cluster Comput 23(4):2609–2627 cloud data centers. Cluster Comput 23(4):3013–3027
Kazim M, Masood R, Awais M (2013) Securing the VM images in Shetty J, Anala R, Shobha G (2012) A survey on techniques of
cloud. In: Proceedings of the 6th International Conference on secure live migration of virtual machine. Int J Comput Appl
Security of Information and Networks 39(12):34–39
Leelipushpam P, Sharmila J (2013) Live VM migration techniques in Thakur S, Kalia A, Thakur J (2013) Server consolidation algorithms
cloud environment—a survey. Info Commun Technol. https://​doi.​ for cloud computing environment: a review. Int J Adv Res Comput
org/​10.​1109/​CICT.​2013.​65581​30 Sci Softw Eng 3(9):379–384
Li Z (2019) An adaptive overload threshold selection process using Wang W, Zhang Y, Lin B, Miao K (2010) Secured VM migration in
Markov decision processes of virtual machine in cloud data center. personal cloud. In: 2nd Intl Conference on Computer Engineer-
Clust Comput 22(2):3821–3833 ing &Tech
Mandal SK, Khilar PM (2013) Efficient virtual machine placement for Wu T, Yang Q, He Y (2017) A secure and rapid response architecture
on-demand access to infrastructure resourcesin cloud computing. for virtual machine migration from an untrusted hypervisor to a
Int J Comput Appl 68(12):6–11 trusted one. Front Comp Sci 11(5):821–835
Medina V, García J (2014) A survey of migration mechanisms of vir-
tual machines. ACM Comput Surv (CSUR) 46(3):1–33 Publisher's Note Springer Nature remains neutral with regard to
Melhem SB, Agarwal A, Goel N, Zaman M (2017) Markov prediction jurisdictional claims in published maps and institutional affiliations.
model for host load detection and VM placement in live migra-
tion. IEEE Access 6:7190–7205 Springer Nature or its licensor holds exclusive rights to this article under
Monil MAH, Rahman RM (2015) Implementation of modified overload a publishing agreement with the author(s) or other rightsholder(s);
detection technique with VM selection strategies based on heuris- author self-archiving of the accepted manuscript version of this article
tics and migration control. In 2015 IEEE/ACIS 14th International is solely governed by the terms of such publishing agreement and
Conference on Computer and Information Science (ICIS), IEEE, applicable law.
pp 223–227
Oberheide J, Cooke E, Jahanian F (2008) Empirical exploitation of
live VM migration. In: Proceeding of BlackHat DC convention,
pp 1–6

13