An energy optimization with improved QOS approach for adaptive cloud resources

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 10, No. 5, October 2020, pp. 4881~4891
ISSN: 2088-8708, DOI: 10.11591/ijece.v10i5.pp4881-4891  4881
Journal homepage: http://ijece.iaescore.com/index.php/IJECE
An energy optimization with improved QOS approach for
adaptive cloud resources
Danthuluri Sudha1
, Sanjay Chitnis2
1
CMR Institute of Technology, Visvesvaraya Technological University (VTU), India
2
Dayanandasagar University, India
Article Info ABSTRACT
Article history:
Received Apr 13, 2019
Revised Mar 6, 2020
Accepted Mar 18, 2020
In recent times, the utilization of cloud computing VMs is extremely
enhanced in our day-to-day life due to the ample utilization of digital
applications, network appliances, portable gadgets, and information
devices etc. In this cloud computing VMs numerous different schemes
can be implemented like multimedia-signal-processing-methods. Thus,
efficient performance of these cloud-computing VMs becomes an obligatory
constraint, precisely for these multimedia-signal-processing-methods.
However, large amount of energy consumption and reduction in efficiency of
these cloud-computing VMs are the key issues faced by different cloud
computing organizations. Therefore, here, we have introduced a dynamic
voltage and frequency scaling (DVFS) based adaptive cloud resource
re-configurability (𝐴𝐶𝑅𝑅) technique for cloud computing devices, which
efficiently reduces energy consumption, as well as perform operations in
very less time. We have demonstrated an efficient resource allocation and
utilization technique to optimize by reducing different costs of the model.
We have also demonstrated efficient energy optimization techniques by
reducing task loads. Our experimental outcomes shows the superiority of our
proposed model 𝐴𝐶𝑅𝑅 in terms of average run time, power consumption and
average power required than any other state-of-art techniques.
Keywords:
ACRR,
Cloud computing,
DVFS,
Energy consumption,
Resource allocation
Copyright © 2020 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Danthuluri Sudha,
CMR Institute of Technology,
Visvesvaraya Technological University (VTU),
Bengaluru, Karnataka 560091, India.
Email: danthulurisudha15@gmail.com
1. INTRODUCTION
Due to ever-enhancing demand and popularity of cloud computing applications, various companies
has moved their focus to cloud computing to decrease costs and for the better utilization of resources hence it
referred as a next generation computing application. Cloud computing application termed as a novel
computational model, which provides on-demand resources and required information, network, storage and
data to the subscribers. The cloud-computing model combines hardware device locations and various
software resources over the cloud network to decrease the management costs. Cloud Computing is a highly
emerged technology which offers high amount of storage capacity, instant scalability and work on
the principle of pay-per-use which is only for the time period subscribers are utilizing it [1]. Cloud
computing applications are distributed into three sections such as Infrastructure-as-a-service (IaaS), Platform
as-a-service (PaaS) and Software as-a-service (SaaS). Virtualization is the most essential technique for cloud
computing applications, which used to decrease resource utilization. Moreover, virtualization helps to active
the numerous virtual machines (VMs) on one machine by allocating every resource, which belongs to
the basic hardware [2].

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 10, No. 5, October 2020 : 4881 - 4891
4882
At the same time, enormous growth of cloud computing applications in recent years has led to rise
of numerous data centers which can led to massive amount of energy consumption. The vast utilization of
electricity led to various environmental issues due to release of huge carbon as well as it enhances
the maintenance cost of information processing centers [3]. The key reason behind massive consumption of
power is the insufficient utilization of resources [4]. For an instance, in a recent report it is concluded that
several researchers have monitored more than 5000 servers for a long 6 months’ time-period and observed
that most of the servers operates only at 10% to 50% of their whole capacity. It is also observed that
generally more number of resources are assigned to VMs than usually required at that time [4]. Moreover,
cloud computing applications experiences extremely variable task-loads due to which server cannot continue
for a long-time period and there is a need of resource re-management course after a specific time-period to
enhance the present resource sharing [3, 5].
Due to modernization in electronic technologies, the performance cost of various devices is highly
emerged as well as the energy consumption in these devices is highly increased [6]. Moreover, several issues
can be occurred such as ecological, economic, and technical issues due to high amount of energy
consumption. Furthermore, extensive power consumption can give rise to cooling costs and if numerous
computing facilities remain active for a long period, then chances of heating problem occurrence can increase
due to which system reliability and availability can increase. A cloud computing application needs support of
compiler time and execution time to run applications precisely. However, due to high-energy consumption,
the execution time of task-loads can be increased from the as usual time.
In cloud computing applications, most essential thing is the allocation of resources efficiently to
increase the performance of cloud information centers. Therefore, to enhance the performance and counter
these issues in recent time different techniques are introduced by various researchers such as Hierarchical
reliability-driven scheduling (HRDS) technique [7], Constrained earliest finish time (CEFT) algorithm [8],
Contention-aware energy-efficient duplication (FastCEED) technique [9], Dynamic voltage and frequency
scaling (DVFS) [10] and voltage and frequency island (VFI) technique [11]. In above all techniques,
Dynamic voltage and frequency scaling (DVFS) is one of the most widely emerged technique for efficient
scheduling of task-loads. DVFS is a highly established energy consumption optimization scheme for
embedded cloud systems and the energy optimization can be achieved by scaling down the voltage of any
chip dynamically. DVFS technique helps to achieve high QoS Internet services by reducing energy
consumption in cloud devices.
However, these existing techniques works upon the classic prototype which is not genuine prototype
and the communication cost of inter processors becomes very high while using these techniques. In [12],
an energy and run-time optimization technique is introduced to decrease energy consumption by scheduling
of various task-loads on numerous embedded systems. In [13], various energy efficient state-of-art-
techniques are discussed and basic requirements of embedded systems are presented to increase
the performance of embedded computing. In [14], a task-scheduling algorithm is presented to achieve trade-
off between energy and performance. This scheduling algorithm helps to increase the flexibility of
heterogeneous computing devices (HCD). This technique attenuates the voltage of various tasks according to
computing model task load. Hence, the power consumption is reduced and a novel interaction aware DAG
prototype is also presented for heterogeneous computing devices (HCD) to reduce energy consumption [15].
However, these techniques do not have features like 𝐷𝑉𝐹𝑆 or Dynamic Network Shutdown (𝐷𝑁𝑆) [16].
Thus, to enhance performance and reduce energy consumption of heterogeneous computing devices
an effective task scheduling technique is needed and rely upon DVFS technique which offers various energy
optimization facilities and helps to achieve better resource utilization and trade-off between performance and
energy. Therefore, we have presented a Dynamic voltage and frequency scaling (DVFS) based adaptive
cloud resource re-configurability (𝐴𝐶𝑅𝑅) technique for cloud computing devices, which efficiently reduces
energy consumption, as well as perform operations in very less time. Our proposed 𝐴𝐶𝑅𝑅 approach helps to
achieve high trade-off between energy and performance in heterogeneous computing devices. This
scheduling technique helps to achieve high 𝑄𝑜𝑆 by operating on lower transmission rates and conceding
minimum delay. This technique supports efficiently to achieve objective of enhancing the cost of
interaction and computational energy. This technique adaptively schedules tasks, allocates its resources, and
provides scalable implantation. Our proposed 𝐴𝐶𝑅𝑅 technique performs much better compare to other
state-of-art-tecniques.
This paper is organized in following sections, which are as follows. In section 2, we present related
work in the field of energy consumption in cloud environment. In section 3, we described our proposed
𝐴𝐶𝑅𝑅 methodology. In section 4, experimental results and performance evaluation shown, and section 5
concludes our paper.

Int J Elec & Comp Eng ISSN: 2088-8708 
An energy optimization with improved QOS approach for adaptive cloud resources (Danthuluri Sudha)
4883
2. RELATED WORK
In recent years, the demand of cloud computing applications has taken immense growth.
Therefore, to control high demand from the clients, there is a need of excessive resources of different types.
All these resources consumes high amount of electricity and hence power consumption is more. According to
a 2013 research in United States of America, the information cloud processing centers consumes
910 𝑏𝑖𝑙𝑙𝑖𝑜𝑛 𝐾𝑊ℎ electricity, which is almost same as the summation of 34 thermal energy plants whose
electricity generation capacity is 500 𝑀𝑊 per year. This consumed energy was sufficient for entire New
York City for two years and till 2020, this energy consumption will rise to 1400 𝑏𝑖𝑙𝑙𝑖𝑜𝑛 𝐾𝑊ℎ which is
enormous electricity consumption and almost same as the summation of 50 thermal energy plants [17].
Hence, the energy consumption in cloud data centers and computing processors has taken drastic growth.
Therefore, controlling of power consumption in data centers and embedded processors is a vital and critical
requirement, which need to be focused soon. Thus, an extensive research work described in this section on
energy balanced scheduling algorithms and their connection with 𝐷𝑉𝐹𝑆 for various embedded devices.
In [18], level of power consumption in information processing centers of china and performance is
measured. They conclude that the power consumption in information processing centers of CHINA is very
high and various techniques are introduced to reduce power consumption and enhance performance. In [19],
an energy consumption model is presented based on the server maximum power and degree of CPU
utilization to predict the total power in the present server. In [20], mobile cloud computing prototype is
introduced to reduce the energy consumption at the time of wireless communication based on dynamic
energy-aware cloudlets. They provide simulations results based on the practical experiments. However,
execution time is very high using this technique, which may degrade its performance. In [21],
an efficient resource allocation model is introduced in cloud environment and a review on existing
scheduling and energy consumption strategies is presented. To offer better resources in cloud environment
and improve relationship with users, scheduling of resources is an extremely essential topic, which can
performance of cloud computing VMs.
In [22], a novel energy aware based on VM scheduling technique is introduced. Here, both network
components and resources both are considered to provide an efficient scheduling technique. VM placement
and VM migration are the two essential scheduling steps to achieve objective. This technique helps to reduce
energy consumption as well as traffic over network. In [23], energy aware resource-scheduling technique is
presented based on DVFS networked information processing centers for cloud computing VMs.
Here, two types of energies are mainly optimized such as computing energy and communication energy to
reduce overall energy consumption while following SLA constraints. This technique is difficult to implement
in real-time. In [24], an efficient cost minimization and resource utilization approach is introduced for cloud
computing devices using stable parallel applications. This approach precisely decreases cost by choosing
devices, which follow the methods of least resource utilization. However, the difficulty is to maintain
trade-off between performance and energy consumption. In [25], a precise scheduling algorithm is introduced
which rely upon DVFS-enabled network processing centers. This algorithm helps to achieve efficient
scheduling for cloud computing VMs. However, this technique introduces optimization problem as well.
In above works, different researchers have utilized different energy consumption and scheduling
techniques. However, only few scheduling techniques are well-known to be offered in real-time applications
due to various problems occurred in techniques [18, 20, 24, 25], like lack of balancing between performance
and power consumption, optimization complexity, large run-time. Therefore, to control these issues,
we have introduced a novel dynamic voltage and frequency scaling (DVFS) based adaptive cloud resource
re-configurability (𝐴𝐶𝑅𝑅) technique for cloud computing devices, which efficiently reduces energy
consumption, as well as perform operations in very less time. Therefore, this technique is very much efficient
to establish a trade-off between performance and energy consumption.
3. PROPOSED ENERGY BALANCED SCHEDULING ARCHITECTURE
This section defines proposed 𝐴𝐶𝑅𝑅architecture and its various modules. This section also describes
about the optimization of computational and re-configuration cost in information processing centers. Figure 1
demonstrates the proposed 𝐴𝐶𝑅𝑅 architecture. Here, we introduce a novel adaptive cloud resource
re-configurability (𝐴𝐶𝑅𝑅) technique for cloud computing VMs. The proposed 𝐴𝐶𝑅𝑅 technique works on
the principle of parallel computing which can handle numerous cloud computing VMs and they can be
controlled by a central resource handler. Every cloud computing VM finishes the present allocated task as
a self-governing processor by self-controlling its memory and resources. Message passing method is used for
Intra-cluster interaction. Whenever a new task is assigned, a central resource handler simultaneously starts
implementing resource distribution and admission governing.

 ISSN: 2088-8708
4884
Figure 1. Architecture diagram of our proposed 𝐴𝐶𝑅𝑅 model
There are three vital components in our proposed 𝐴𝐶𝑅𝑅 technique which helps to achieve better
resource utilization from an infrastructure perception such as information storage, switched local area
network (LAN) and virtual machine handler (VMC) as demonstrated in Figure 1. Whenever, a new task is
assigned the arrival time of that task is defined by 𝑎 𝑚 and size of that task is denoted by 𝑆 𝑎 in bits. The total
processing time of assigned task is less than or equal to the estimated employed time 𝐸 𝑎(𝑠𝑒𝑐)which is very
essential for any technique to be adopt in real time scenarios. Our proposed 𝐴𝐶𝑅𝑅 technique works on some
essential parameters which is necessary to work in real time scenarios such as processing task size 𝑆 𝑎,
the maximum allowed delay 𝐸 𝑎 in sec and the task granularity which shows the maximum number of tasks
(𝑁𝐸 ≫ 1) can be grouped into the assigned work.
Assume that the maximum VMs, which can be utilized in the assigned tasks using our proposed
𝐴𝐶𝑅𝑅methods, can be expressed by 𝑁𝐽 ≫ 1and presented in Figure 1. Our scheduling technique works on
the principle that every VM can be demonstrated as a virtual server, which can process 𝑖 𝑔 𝑏𝑡𝑠 (Bits per
Second). The operating rate 𝑖 𝑔can be parallelly scaled at the time of execution depending upon the task size
𝑆 in bits. Assume that all the task follows the interval [0, 𝑖 𝑔
↑
] where 𝑖 𝑔
↑
belongs to the maximum permissible
operating rate.
Moreover, the task size 𝑆 does not affect the estimated time to complete that assigned task by VM,
which is fixed prior only to be adopt our model in real time scenarios and denoted by 𝑀 in seconds.
Furthermore, a VM can handle background task-loads of a present assigned task whose size is 𝑆 and
the background task-load size is𝑆 𝕓 . This background task-load comes under OS (operating system)
programs. It is assumed that the background task-load is stored by basic memory of VM. Thus,
the background task-load only required computing cost and does not persuade interaction cost. Thus,
the utilization parameter 𝛽 can be expressed as,
𝛽 ≜ 𝑖 𝑔 . (𝑖 𝑔
↑
)
−1
∈ [0,1], (1)
where, (1) represents that the dynamic elements of the computing energy are the most essential part to
decrease the computational cost. Assume that the total energy consumption by VM is to finish a single task
of time interval 𝜔(𝕕)is denoted by 𝜑 𝑔 in joule at the operating rate𝑖 𝑔. Thus, the dimensionless ratio can be
expressed as,
𝜃(𝛽) ≜ 𝜑 𝑔(𝑖 𝑔). (𝜑 𝑔
↑
)
−1
≡ 𝜃 (𝑖 𝑔). (𝑖 𝑔
↑
)
−1
, (2)
Virtual machine controller
and task scheduler
Switch
CPU, Servers and storage devices
Basic layer (Virtualization)
…
….
.
…
…
To the
internet
From
the
internet
Output
task load
Input task
load
Gateway
Router
Vm(1) Vm(2) Vm(m)

4885
where, (2) represents the Total Energy Consumption by the concern VM. For an instance, the DVFS based
CPU analytical form can be described by the following equation,
𝜃(𝛽) = 𝛽2
, 𝛽 𝜖 [0,1], (3)
Here, we can also use 𝜇 to compute relative energy cost by concerned VM for the completion of task.
3.1. Modelling for task-load reduction using proposed 𝑨𝑪𝑹𝑹 technique
In this section, modelling for task-load reduction is discussed. Assume that 𝑁 ≜ ↓ {𝑁𝐽 , 𝑁𝐸} is
the number of tasks which are not overlapped and can be performed in parallel to execute various tasks.
Assume that 𝑆 𝑑is the task size which are assigned to the computing 𝑉𝑀 (𝑑). The process time of different
tasks does not rely upon the task length𝑆 𝑑. Therefore, the processing rate can be defined as in bits
per seconds,
𝑖 𝑔(𝑑) = 𝑆 𝑑 . (𝜔)−1
, (4)
This (4) shows that the maximum length permitted for a task is 𝑆 𝑑
↑
= 𝜔. 𝑖 𝑔
↑(𝑑). And
𝛽 ≜ 𝑖 𝑔(𝑑). (𝑖 𝑔
↑
)
−1
≡ 𝑆 𝑑. (𝑆 𝑑
↑
)
−1
. The total size of a job can be referred as 𝑆 𝑎 in bits and task size of
the task, which is assigned to the 𝑉𝑀(𝑑)by task scheduler as shown in Figure 1, can be referred as
𝑆 𝑑 ≫ 0, 𝑑 = 1, … . , . . 𝑁. To reduce task loads, we distribute total job size 𝑆 𝑎 into 𝑁 parallel tasks whose size
boundary limit can be defined as∑ 𝑆 𝑑 = 𝑆 𝑎
𝑁
𝑑 .
3.2. Optimization of reconfiguration cost using proposed 𝑨𝑪𝑹𝑹 technique
This section provides detailed modelling for the optimization of reconfiguration cost. The VM
module controller is used to perform two key operations such as balancing the task loads and controlling of
virtual machines. For the controlling of virtualization layer as demonstrated in Figure 1, the virtual machine
controller (𝑉𝑀𝐶)is required which helps to achieve final mapping of VM resources on numerous computing
VMs. The VM’s characteristics parameters can be described by (9) as,
{𝜔, 𝑖 𝑔
↑(𝑑), 𝜃 𝑑(𝛽 𝑑), 𝜑 𝑔
↑ (𝑑), 𝜇(𝑑), 𝑆 𝕓(𝑑), 𝑑 = 1, … … . . 𝑁 } (5)
where, all these parameters can be stated using Virtualization Layer and then they are transmitted to virtual
machine controller (𝑉𝑀𝐶) as demonstrated in Figure 1. The operating rate 𝑖 𝑔 can be scaled up or scaled
down using an efficient frequency- scaling scheme, which is controlled by 𝑉𝑀𝐶. The power consumption
while switching from operating frequency 𝑖1 to frequency 𝑖2 can be 𝜑(𝑖1: 𝑖2) in joule. This power
consumption mainly rely upon the technique used and on the CPU’s present in the workstation. This function
𝜑(𝑖1: 𝑖2) consists of some properties such as the function 𝜑(𝑖1: 𝑖2) rely upon the entire frequency gap
|𝑖1 − 𝑖2|, it becomes zero at 𝑖1 = 𝑖2and remain non-decreasing in the entire frequency gap|𝑖1 − 𝑖2|, it is
combined convexly at 𝑖1 and𝑖2. Our model 𝐴𝐶𝑅𝑅 have some characteristics which can be shown using (10),
𝜑(𝑖1: 𝑖2) = 𝕖 𝑓(𝑖1 − 𝑖2)2
, 𝑗𝑜𝑢𝑙𝑒 (6)
where, 𝕖 𝑓 represents reconfiguration cost for the unit switching of frequency and the values of 𝕖 𝑓 is bounded
only to some hundreds of 𝜇𝑗𝑜𝑢𝑙𝑒𝑠/(𝑀𝐻𝑧)2
. In our model 𝐴𝐶𝑅𝑅, for every job the size 𝑆 𝑎 remains same over
the respective operating time 𝐸 𝑎 and any kind of fluctuations not occurred in the task-loads during task-
execution. Various tasks can be parallel executed at run-time, due to the induced time overhead using
frequency-scaling technique is very less in few 𝜇𝑠𝑒𝑐 for DVFS-enabled architectures. The above-mentioned
prediction that the utilization parameter 𝛽 can be continuous valued and it requires continuous computational
rates, which is denoted by𝑖 𝑔. The 𝑉𝑀𝑠 can offer an instance of CPU’s which offer a finite set as,
ℍ ≜ {𝑖̂(0)
≡ 0, 𝑖̂(1)
, … , 𝑖̂(𝕃−1)
≡ 𝑖 𝑔
↑
}, (7)
where, these finite set consists of discrete computational rate 𝕃. The optimality loss from both continuous and
discrete DVFS enabled techniques, can be eliminated by (8) as,
𝔽 ≜ {𝛽̂(0)
≡ 0, 𝛽̂(1)
, … … , 𝛽̂(𝕃−1)
≡ 1}, (8)

 ISSN: 2088-8708
4886
where, the discrete value set of 𝛽 which represent the frequency set ℍ as shown in (7). A virtual power
consumption curve can be denoted as 𝜃̃(𝛽) and formed using piecewise linear interpolation and the
permitted operating points are,
{(𝛽̂(𝕠)
, 𝜃 (𝛽̂(𝕠)
)) , 𝕠 = 0, … … . . , (𝕃) − 1} (9)
where, the corresponding vertex points can be presented as,
(𝛽̂(𝕠)
, 𝜃 (𝛽̂(𝕠)
)) 𝑎𝑛𝑑 (𝛽̂(𝕠+1)
, 𝜃 (𝛽̂(𝕠+1)
)) (10)
This above mentioned 𝑣𝑖𝑟𝑡𝑢𝑎𝑙 𝑝𝑜𝑤𝑒𝑟 𝑐𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛 𝑐𝑢𝑟𝑣𝑒 maintains the continuity and can be
used for the provisioning of 𝑉𝑀 resources. The use of 𝑝𝑖𝑒𝑐𝑒𝑤𝑖𝑠𝑒 𝑙𝑖𝑛𝑒𝑎𝑟 𝑖𝑛𝑡𝑒𝑟𝑝𝑜𝑙𝑎𝑡𝑖𝑜𝑛 suggests that with
the help of 𝑣𝑖𝑟𝑡𝑢𝑎𝑙 𝑝𝑜𝑤𝑒𝑟 𝑐𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛 𝑐𝑢𝑟𝑣𝑒, the average energy cost of DVFS enabled techniques
remains under the estimated interval of time duration 𝜔. Here, every 𝑉𝑀 configuration rely upon CPU type,
size of memory and cost per time. The 𝑉𝑀𝑠 cost rely upon the type of configuration. The internal cost of
𝑉𝑀𝑠 assumed to be zero in all information cloud centers.
3.3. Modelling of efficient resource allocation
In this section, modelling for effective resource allocations is presented. Here, 𝑉𝑀𝐶 offers two types
of services such as balancing of the load and sharing of computational resources. Precisely, these service
used to fine-tune the rate of computation (𝑖 𝑑 , 𝑑 = 1, … … … . , 𝑁)and size of task (𝑆 𝑑 , 𝑑 = 1, … … … . , 𝑁)
for the DVFS enabled cloud computing 𝑉𝑀𝑠 as demonstrated in Figure 1. The main objective is to reduce
the total computational energy in joule, which is defined in (11),
𝜑 𝕋 ≜ ∑ 𝜑 𝑔(𝑑)𝑁
𝑑=1 , (11)
where, the total computational energy consumption 𝜑 𝕋 rely on the run-time/job 𝐸 𝑎 in seconds, operating
time/task (𝜔) needed by 𝑉𝑀𝑠 demonstrated in Figure 1. Precisely, in Figure 1 all the 𝑁 links are operated by
the switching unit adaptively. The entire computational overhead for the 𝑑 𝑡ℎ
link can be expressed as,
2𝑀(𝑑) + 𝜔, (12)
where, the condition on total run-time per job to satisfy the solution of optimization problem can be
expressed as,
max
1≤𝑑≤𝑁
{2𝑀(𝑑)} + 𝜔 ≤ 𝐸 𝑎, (13)
Assume that the total computational energy optimization issues can be expressed as following,
min
{ℂ 𝑑,𝑖 𝑑,𝑆 𝑑}
∑ 𝜃 𝑑 (𝑖 𝑑. (𝑖 𝑑
↑
)
−1
) 𝜇(𝑑)𝑁
𝑑=1 𝜑 𝑑
↑
+ 𝕖 𝑓(𝑖 𝑑 − 𝑖 𝑑
𝑛)2
, (14)
where it states that,
(𝑆 𝑑 + 𝑆 𝕓(𝑑)) ≤ 𝑖 𝑑 𝜔, 𝑑 = 1, … … … 𝑁, (15)
∑ 𝑆 𝑑 = 𝑆 𝑎,𝑁
𝑑=1 (16)
0 ≤ 𝑖 𝑑 ≤ 𝑖 𝑑
↑
, 𝑑 = 1, … … … 𝑁, (17)
𝑆 𝑑 ≥ 0, 𝑖 = 1, … … . , 𝑁, (18)
where, in (14) starting first term represents computational energy and the following term represents
re-configuration energy, which can be expressed by 𝜑 𝑔(𝑑) jointly by the computing VMs. Moreover, in (14),
𝑖 𝑑
𝑛
represents the present rate of computation and 𝑖 𝑑 is the required rate of computation. Here, 𝑖 𝑑
𝑛
remains
constant while processing of task and shows present state of 𝑉𝑀(𝑑) whereas 𝑖 𝑑 can be variable and changes

4887
task to task. Here, (15) shows the condition for which assigned task must be executes in 𝜔 seconds whereas
(16) shows the condition in which the assigned job must be divided into 𝑁 parallel tasks.
3.4. Solution for optimization problem using 𝑨𝑪𝑹𝑹
This section provide various solution for handling optimization problem using our efficient
scheduling architecture 𝐴𝐶𝑅𝑅. Firstly, let the time delay occurred while switching of frequencies remains
constant which is induced by DVFS-enabled-techniques. However, the time delay occurred while switching
of frequencies between various 𝑉𝑀𝑠 can be optimized by a non-negative function𝜌 𝑑(𝑖 𝑑, 𝑖 𝑑
𝑛)(𝕕) and
the solution for optimization problem, which is stated in (15) and (18), can be derived. This non-negative
function𝜌 𝑑(𝑖 𝑑, 𝑖 𝑑
𝑛)(𝕕) helps to retain two characteristics such as,
a. The non-negative function𝜌 𝑑(. , . ) remains non-decreasing throughout the interval|𝑖 𝑑 − 𝑖 𝑑
𝑛|.
b. The product of 𝑖 𝑑 and non-negative function𝜌 𝑑(. , . ) remain convex in𝑖 𝑑.
To control optimization problem, the first term in (14) which represent computational energy
∑ 𝜃 𝑑 (𝑖 𝑑. (𝑖 𝑑
↑
)
−1
) . 𝜇(𝑑)𝑁
𝑑=1 . 𝜑 𝑑
↑
can be interchanged by the energy function 𝔾(𝑖1, … … … . , 𝑖 𝑁). Similarly,
the re-configuration energy 𝕖 𝑓(𝑖 𝑑 − 𝑖 𝑑
𝑛)2
can be interchanged by the energy function 𝐽(𝑖1, … … … . 𝑖 𝑁) which
are conjointly convex in the interval{𝑖 𝑑, 𝑑 = 1, … … … . . 𝑁}2
. Similarly, in (15) the parameter 𝜔 can be
interchanged by 𝜔(𝑑) without changing any condition or linear equation. The optimization problem
presented from (14) to (18) is not a convex type. In fact, this problem is a loosely coupled optimization type,
where the parameter 𝑆 𝑑, 𝑑 = 1, … … … 𝑁 represents the computational problems. The solution for
optimization of computational and re-configurational problem is presented as,
min
{𝑆 𝑑,𝑖 𝑑}
∑ (𝜃 𝑑 (𝑖 𝑑. (𝑖 𝑑
↑
)
−1
) 𝜇(𝑑)𝜑 𝑑
↑
+ 𝕖 𝑓(𝑖 𝑑 − 𝑖 𝑑
𝑛)2
)𝑁
𝑑=1 (19)
Assume that, {𝑆 𝑑
∗
, 𝑖 𝑑
∗
, 𝑑 = 1, … … … 𝑁} represents the solution set for computational and
re-configurational optimization problem, which is shown in (15) to (18). At last, we present our proposed
𝑃𝐸𝐵𝑆 model in an efficient algorithm form, which is as follows:
Algorithm : 𝑨𝑪𝑹𝑹 algorithm
1. Fix (𝑁𝐽, 𝐸, 𝐸 𝑎)
2. Fix (𝑖 𝑑, 𝜇(𝑑), 𝑖1
↑
, 𝑆 𝑎)
3. Fix (𝜑 𝑔, 𝑆 𝑎, 𝜌 𝑑, 𝜃(𝛽) , 𝛽, )Parameter for channel processing of 𝑁𝐽 VMs
4. Collect 𝑆 𝑑
5. Verify the attainable constraints of (14)
6. 𝕚1 ≅ (𝑆 𝑑 + 𝑆 𝕓(𝑑)) ≤ 𝑖 𝑑 𝜔, 𝑑 = 1, … … … 𝑁,
7. 𝕚2 ≅ ∑ 𝑆 𝑑 = 𝑆 𝑎,𝑁
𝑑=1
8. if ≈ (𝕚1&𝕚2) 𝑡ℎ𝑒𝑛,
9. 𝑒𝑟𝑟𝑜𝑟(𝑝𝑟𝑜𝑔𝑟𝑎𝑚 𝑖𝑠 𝑛𝑜𝑡 𝑎𝑡𝑎𝑖𝑛𝑎𝑏𝑙𝑒)
10. else
11. Special optimization complexity :
12. min (𝜑 𝕋, 𝕖𝑓, 𝜑 𝑔(𝑑))
13. subjected to:
14. conditions in (15) and (16)
15. end if
16. return 𝜑 𝕋, 𝕖𝑓, 𝜑 𝑔(𝑑)
basic SLA constraints
Parameter processing of VMs 𝑁𝐽
4. PERFORMANCE EVALUATION
Now days, the request of cloud computing devices has highly emerged in real-time due to
the extensive utilization of informative devices, digital instruments, network appliances and portable gadgets
etc. Multimedia-signal-processing method is well-known technique, which can be utilized in these
cloud-computing devices. Therefore, the performance of these computing devices must be superior due to
the extensive demand of these computing devices in day-to-day life. However, high-energy consumption in
these computing devices can disturb their performance. Thus, this section discusses about the balancing
between performance and power consumption. To achieve these objectives, we have introduced a Dynamic
voltage and Frequency Scaling (DVFS) based Adaptive Cloud Resource Re-Configurability (𝐴𝐶𝑅𝑅)
technique for heterogeneous computing devices, which efficiently reduces energy consumption, as well as
provide superior performance. The run-time can be evaluated considering various jobs as 30, 50, 100,
and 1000. Graphical representation of our outcomes is also presented considering execution time, number of
tasks and energy consumption. The run-time and total power consumed can be evaluated using different
parameters in Table 1, which is demonstrated in the following section. Our proposed 𝐴𝐶𝑅𝑅 model is tested

 ISSN: 2088-8708
4888
on 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 scientific dataset. We have considered different sizes of scientific workflow experiments as
30, 50, 100 and 1000. Our proposed 𝐴𝐶𝑅𝑅model employed on 64-bit windows 10 OS with 16 GB RAM
which consists on INTEL (R) core (TM) i5-4460 processor. It consists of 3.20 GHz CPU. This project is
simulated using 𝐸𝑐𝑙𝑖𝑝𝑠𝑒𝑊𝑆 Neon.3 editor and code is written in JAVA.
4.1. Comparative study
In this modern era, computing devices has ruled market in different fields like medical, healthcare
solutions, trading, software companies etc. Thus, future expertise is clearly in favor of these cloud-computing
devices due to their extensive requirements. However, the efficiency of these computing devices may be
reduced due to high-energy consumption and lack of efficient resource utilization techniques. Consequently,
these issues can be sorted out using efficient task scheduling techniques. Therefore, to allocate resource
properly and schedule all the tasks efficiently to overcome power consumption problem, we have presented
a novel Dynamic voltage and Frequency Scaling (DVFS) based Adaptive Cloud Resource Re-Configurability
(𝐴𝐶𝑅𝑅) technique. A precise task scheduling technique can enhance throughput of the system, increase
connections with subscribers, offer better resource utilization and can afford to handle multiple tasks at
a time etc. The results are demonstrated in contrast to other state-of-art techniques in terms of energy
consumption, run-time, power sum and average power as shown in Table 1 using scientific model
CyberShake for various jobs as 30, 50,100 and 1000. Energy consumption using our proposed 𝐴𝐶𝑅𝑅
technique for CyberShake 30 is 1303.74 Watts, CyberShake 50 is 1330.92 Watts,CyberShake 100 is
1436.83 Watts and CyberShake 1000 is 3228.60 Watts demonstrated in Table 1, which is very less in
contrast to other state-of-art techniques. Table 2 demonstrates Average Execution time evaluation using our
𝐴𝐶𝑅𝑅 technique and comparison with other state-of-art-techniques presented for scientific model
CyberShake for various jobs as 30, 50,100 and 1000. The Average Execution Time using our 𝐴𝐶𝑅𝑅
technique for scientific model CyberShake 100 is 31.162 sec and CyberShake 1000 is 4.974 sec.
Table 1. Various parameters comparison for proposed ACRR technique vs DVFS
using scientific model 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒
Parameters
DVFS 𝐴𝐶𝑅𝑅
𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒
30
50
100
1000
30
50
100
1000
VM=20 VM=30 VM=60 VM=20 VM=20 VM=30 VM=60 VM=20
Total
Execution
Time (s)
6359.41 14448.90 30124.41 74543.57 2938.22 2953.86 3116.22 4794.39
Power Sum
(W)
12175922.64 29068552.89 61177338.40 149968122.08 4646390.51 4696963.35 4955133.77 7623610.31
Average
Power (W)
19.146320 20.118183 20.308229 20.118184 15.813623 15.901103 15.901103 15.901105
Power
Consumption
(𝑊ℎ)
3495.42 8518.39 18966.33 236303.28 1303.74 1330.92 1436.83 3228.60
Table 2. Average execution time comparison of proposed 𝐴𝐶𝑅𝑅 technique with
other State-of-art-techniques using scientific model 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒
DAGs
Number of
nodes
Average Execution time (s)
EMO-based algorithm [26] DVFS 𝐴𝐶𝑅𝑅
𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 100 100 31.53 301.244 31.162
𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 1000 1000 22.71 74.54 4.974
4.2. Graphical representation
This section demonstrates the graphical representation of our evaluated outcomes. Here, Figure 2
shows run time comparison of our proposed 𝐴𝐶𝑅𝑅 technique with DVFS technique using scientific workload
𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒for different jobs as 30, 50, 100 and 1000. Here, Figure 3 shows Power Sum Comparison of our
proposed 𝐴𝐶𝑅𝑅 technique with DVFS technique using scientific workload 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 for different jobs as
30, 50, 100 and 1000. Here, Figure 4 shows Average Power Required Comparison of our proposed
𝐴𝐶𝑅𝑅 technique with DVFS technique using scientific workload 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 for different jobs as 30, 50,
100 and 1000. Similarly, Figure 5 shows Power Consumption Comparison of our proposed 𝐴𝐶𝑅𝑅 technique
with DVFS technique using scientific workload 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 for different jobs as 30, 50, 100 and 1000.
Furthermore, Figure 6 shows Average Run Time Comparison of our proposed 𝐴𝐶𝑅𝑅 technique with DVFS

4889
technique using scientific workload 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 for different jobs as 30, 50, 100 and 1000. This result
demonstrates the superiority of our proposed 𝐴𝐶𝑅𝑅 technique in terms of power consumption, execution
time, average power and time.
Figure 2. Run-time comparison using our ACRR technique with DVFS
Figure 3. Power sum comparison using our ACRR technique with DVFS
Figure 4. Average power comparison using our ACRR technique with DVFS
0
20000
40000
60000
80000
VM=20 VM=30 VM=50 VM=30
CYBERSHAKE 30 CYBERSHAKE 50 CYBERSHAKE 100 CYBERSHAKE 1000
Run-time comparison of our ACRR technique with DVFS
DVFS (s) ACRR (s)
0
20000000
40000000
60000000
80000000
100000000
120000000
140000000
160000000
VM=20 VM=30 VM=50 VM=30
Power Sum comparsion of our ACRR technique with DVFS
DVFS (W) ACRR (W)
0
5
10
15
20
25
VM=20 VM=30 VM=50 VM=30
Average Power comparsion of our ACRR technique with DVFS
DVFS (W) ACRR (W)

 ISSN: 2088-8708
4890
Figure 5. Power consumption comparison using our ACRR technique with DVFS
Figure 6. Average execution time comparison using our ACRR technique with DVFS
5. CONCLUSION
The significance of controlling high-energy consumption and task load allocation for every cloud
computing VMs is very essential. Therefore, to attain balancing between power consumption and
performance for computing processors in cloud environment, we have introduced a novel dynamic voltage
and frequency scaling (DVFS) based adaptive cloud resource re-configurability (𝐴𝐶𝑅𝑅) technique for cloud
computing VMs. Efficient modelling for all the three types of cost such as computation cost and
re-configuration cost is presented. The performance of the model is enhanced by reducing all three costs.
Furthermore, modelling for efficient resource allocation and utilization is presented and task loads are
reduced which is a great challenge for other state-of-art techniques. Numerous resources can be efficiently
allocated adaptively at a time using this technique. The experimental results are shown in terms of run time
taken, reduction in energy consumption and average power required for cloud computing VMs. The Average
Run Time using our proposed 𝐴𝐶𝑅𝑅 model for scientific model 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 100 is 31.162 sec and
𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 1000 is 4.974 sec. Similarly, Power Consumption for 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 30 is 1303.74 Watts,
𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 50 is 1330.92 Watts, 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 100 is 1436.83 Watts and 𝐶𝑦𝑏𝑒𝑟𝑆ℎ𝑎𝑘𝑒 1000 is 3228.60
Watts which is very low compare to other state-of-art-techniques. Our experimental results verifies
the superiority of our model 𝐴𝐶𝑅𝑅 in terms of performance and power consumption in contrast to other
state-of-art-techniques.Therefore, a trade-off between performance and energy consumption is maintained
using our proposed 𝐴𝐶𝑅𝑅 algorithm. In future, efficient modelling to optimize communication cost will
be presented.
0
50000
100000
150000
200000
250000
VM=20 VM=30 VM=50 VM=30
Power Consumption comparsion of our ACRR technique with DVFS
DVFS (Wh) ACRR (Wh)
0
100
200
300
400
EMO DVFS PS1
Average Execution Time comparsion of our ACRR technique with DVFS
CYBERSHAKE 100 CYBERSHAKE 1000

4891
REFERENCES
[1] T. Zaidi and R. Rampratap, “Virtual machine allocation policy in cloud computing environment using CloudSim,”
International Journal of Electrical and Computer Engineering (IJECE), vol. 8, no. 1, pp. 344-354, 2018.
[2] Kumaraswamy S. and M. K. Nair, “Bin packing algorithms for virtual machine placement incloud computing: a
review,” International Journal of Electrical and Computer Engineering (IJECE), vol. 9, no. 1, pp. 512-524, 2019.
[3] A. Orgerie, et al., “A Survey on Techniques for Improving the Energy Efficiency of Large-Scale Distributed
Systems,” ACM Comput. Surveys, vol. 46, no. 4, 2014.
[4] A. Beloglazov, et al., “A taxonomy and survey of energy-efficient data centers and cloud computing systems,”
Advances in Computers, vol. 82, pp/ 47-111, 2011.
[5] A. Hameed, et al., “A survey and taxonomy on energy efficient resource allocation techniques for cloud computing
systems,” Springer-Verlag Wien, 2014.
[6] Wikipedia, “Moore’s law,” 2012. [Online]. Available: http://en.wikipedia.org/wiki/Moore’s_law.
[7] X. Tang, et al., “A hierarchical reliability-driven scheduling algorithm in grid systems,” Journal of Parallel and
Distributed Computing, vol. 72, no. 4, pp. 525-535, 2012.
[8] M. A. Khan, “Scheduling for heterogeneous systems using constrained critical paths,” Parallel Comput, vol. 38,
no. 4, pp. 175-193, 2012.
[9] J. Singh, et al., “Contention aware energy efficient scheduling on heterogeneous multiprocessors,” IEEE
Transactions on Parallel and Distributed Systems, vol. 26, no. 5, pp. 1251-1264, May 2015.
[10] L. Zhang, et al., “Contention-Aware Reliability Efficient Scheduling on Heterogeneous Computing Systems,”
in IEEE Transactions on Sustainable Computing, vol. 3, no. 3, pp. 182-194, 2018.
[11] J. Kim and M. Park, “Dynamic frequency scaling regarding memory for energy efficiency of embedded systems,”
International Journal of Electrical and Computer Engineering (IJECE), vol. 8, no. 3, pp. 1798-1804, 2018.
[12] K. Li, “Energy-efficient task scheduling on multiple heterogeneous computers: Algorithms, analysis, and
performance evaluation,” IEEE Transactions on Sustainable Computing, vol. 1, no. 1, pp. 7-19, 2016.
[13] A. Munir, et al., “High-performance energy-efficient multicore embedded computing,” IEEE Transactions on
Parallel and Distributed Systems, vol. 23, no. 4, pp. 684-700, 2012.
[14] X. Zhu, et al., “3e: Energy-efficient elastic scheduling for independent tasks in heterogeneous computing systems,”
Journal of Systems and Software, vol. 86, no. 2, pp. 302-314, 2013.
[15] D. Kliazovich, et al., “Ca-dag: Modeling communication-aware applications for scheduling in cloud computing,”
Journal of Grid Computing, vol. 14, no. 1, pp. 23-39, 2015.
[16] R. N. Calheiros, et al., “CloudSim: a toolkit for modeling and simulation of cloud computing environments and
evaluation of resource provisioning algorithms,” Software Practice and Experience, vol. 41, no. 1, pp. 23-50, 2011.
[17] Peter K.D., “Mind the Gap: Energy Availability and the Disconnect with Data,” Data-Center-Issue-Paper-final826,
[Online]. Available: https://verneglobal.com/uploads/VG-Mind-the-Gap-Energy-Availability-and-the-Disconnect-
with-Data-2016.pdf
[18] L. J., et al., “Research on national energy consumption and efficiency of data center,” Energy of China, vol. 11,
pp. 42-45, 2010.
[19] A. Belogazov, et al., “Energy-aware resource allocation heuristics for efficient management of data centers for
Cloud computing,” Future Generation Computer Systems, vol. 28, no. 5, pp. 755-768, 2012.
[20] K. Gai, et al., “Dynamic energy-aware cloudlet-based mobile cloud computing model for green computing,”
Journal of Network & Computer Applications, vol. 59, pp. 46-54, 2016.
[21] S. Selvarani, et al., “Enhanced two stage heuristics algorithm for VM scheduling,” 2016 Inter. Conf. on Advanced
Communication Control and Computing Technologies (ICACCCT), Ramanathapuram, pp. 726-729, 2016.
[22] E. I. Nehru, et al., “Two stage optimal VM Scheduling scheme for IaaS Cloud,” in Proc. International Conference
on Communication and Security, Pondicheerry Engineering College, 2016.
[23] N. Kaur, et al., “Towards energy efficient scheduling with DVFS for precedence constrained tasks on
heterogeneous cluster system,” 2015 2nd
International Conference on Recent Advances in Engineering &
Computational Sciences (RAECS), Chandigarh, pp. 1-6, 2015.
[24] G. Xie, et al., “Resource consumption cost minimization of reliable parallel applications on heterogeneous
embedded systems,” in IEEE Transactions on Industrial Informatics, vol. 13, no. 4, pp. 1629-1640, Aug 2017.
[25] K. A. Sultanpure, et al., “An Efficient Cloud Scheduling Algorithm For The Conservation Of Energy
Through Broadcasting,” International Journal of Electrical and Computer Engineering (IJECE), vol. 8, no. 1,
pp. 179-188, 2018.
[26] Z. Zhu, et al., “Evolutionary multi-objective workflow scheduling in cloud,” in IEEE Transactions on Parallel and
Distributed Systems, vol. 27, no. 5, pp. 1344-1357, May 2016.

An energy optimization with improved QOS approach for adaptive cloud resources

More Related Content

What's hot (17)

Similar to An energy optimization with improved QOS approach for adaptive cloud resources (20)

More from IJECEIAES (20)

Recently uploaded (20)

An energy optimization with improved QOS approach for adaptive cloud resources