REAL-TIME ADAPTIVE ENERGY-SCHEDULING ALGORITHM FOR VIRTUALIZED CLOUD COMPUTING

International Journal of Distributed and Parallel Systems (IJDPS) Vol.10, No.1, January 2019
DOI:10.5121/ijdps.2019.10101 1
REAL-TIME ADAPTIVE ENERGY-SCHEDULING
ALGORITHM FOR VIRTUALIZED CLOUD COMPUTING
D B Srinivas1
, H K Krishnappa2
, Rajan M A3,
Sujay N. Hegde4
1,4
Nitte Meenakshi Institute of Technology, Bangalore, India
2
R V College of Engineering, Bangalore, India
3
TCS Research and Innovation, Bangalore, India
ABSTRACT
Cloud computing becomes an ideal computing paradigm for scientific and commercial applications. The
increased availability of the cloud models and allied developing models creates easier computing cloud
environment. Energy consumption and effective energy management are the two important challenges in
virtualized computing platforms. Energy consumption can be minimized by allocating computationally
intensive tasks to a resource at a suitable frequency. An optimal Dynamic Voltage and Frequency Scaling
(DVFS) based strategy of task allocation can minimize the overall consumption of energy and meet the
required QoS. However, they do not control the internal and external switching to server frequencies,
which causes the degradation of performance. In this paper, we propose the Real Time Adaptive Energy-
Scheduling (RTAES) algorithm by manipulating the reconfiguring proficiency of Cloud Computing-
Virtualized Data Centers (CCVDCs) for computationally intensive applications. The RTAES algorithm
minimizes consumption of energy and time during computation, reconfiguration and communication. Our
proposed model confirms the effectiveness of its implementation, scalability, power consumption and
execution time with respect to other existing approaches.
KEYWORDS
Virtual Machine (VM); Virtualized Data Centers (VDCs); Quality of Service (QoS); Dynamic Voltage and
Frequency Scaling (DVFS); Real Time Adaptive Energy-Scheduling (RTAES).
1. INTRODUCTION
Increased availability of the cloud models and allied developing models allows cloud service
providers to provide easier and cost effective services to the cloud users. The cloud computing
model generally depends upon virtualization technology, which allows multiple operating
systems (OSs) to share a hardware system in an easy, secure and manageable way.
As more and more services are digitized and added to the cloud, cloud users have started using
cloud services. Thus, the computing resources in cloud consume significant amount of power to
serve user requests. This is evident from the following facts. Middle size data centers consume
80000 kW [1]. It is predicted that computing resources consumes nearly 0.5% of the world’s total
power usage [2], and if current demand continues, this is projected to quadruple by 2020. From
these observations, it is evident that optimizing energy consumption is an important issue for
cloud services in a data center. Energy minimization can be done in two ways (i) Designing
energy-efficient processors. (ii) Minimizing the energy consumption by task scheduling.
To reduce energy consumption in the cloud environment, a Dynamic Voltage and Frequency
Scaling (DVFS) approach is used. A DVFS is the hardware model, that has been used to increase

2
and decrease the frequency of CPUs depending upon the user’s requirements [3] and is widely
used in modern processors. The optimistic correlation of energy consumption and the CPU
frequency can control the energy consumption. The use of a DVFS in a processor can
dynamically change the working frequency and leads to a diverse consumption of energy. A
lower frequency that consumes less energy is always not correct because a low frequency that
maximizes the execution time of a task; also depends upon both the execution time and the
power. In [4, 5], the authors specify that there are some optimum frequencies for a given
processor, where the consumption of energy can reduced during the execution of a task.
In a data center, operating a processor at an optimal frequency results in minimum energy
consumption. Generally user application contains set of tasks. If all tasks have been executed
under the optimum frequency then there is a chance that some tasks may not meet their deadline.
Therefore, efficient task allocation with a suitable frequency is a major challenge, since the
optimum suitable frequency for each processor can optimize the total consumption of energy. An
optimal DVFS-based strategy for task allocation should minimize the overall consumption of
energy and meeting the required QoS (i.e., perform the task on time). A cost-constrained
Heterogeneous Earliest Finish-Time (HEFT) model was studied in [6, 7], and these algorithms
provides the optimum budget in different scenarios. The execution cost minimization for
scientific workflows with a deadline constraint is considered and the meta-heuristic algorithm is
proposed in [8]. The investigation of servers’ power consumption with different numbers of VMs
and various utilization has discussed in [9]. A previous study of the ‘worst case of execution
cycle/time’ (WCEC/WCET) was presented in [10-12], where the execution time of the upper
bound at the maximal frequency acts as the task workload. However, the WCET generally does
not contest the running workload of a real task, which may result in a difference between the real
energy savings and the theoretical results. Similarly in [13, 14] have made use of the probability
function based on WCET in a workload to enhance the effectiveness of scheduling tasks.
Assessing the energy utilized by Virtualized Data Centers (VDCs) is one emerging technique that
aims to implement an effective energy management for a virtualized computing platform [15, 16].
The main aim is to deliver high-class service to large number of clients, while minimizing the
overall networking and computing energy consumption. In paper [17], they considered the power
management and optimal resource allocation in a VDC switch with heterogeneous applications;
but they did not control for the internal and external switching to server frequencies, which causes
degrades the performance. In this paper, we propose the RTAES algorithm by manipulating the
reconfiguration proficiency of Cloud Computing-Virtualized Data Centers (CCVDCs) that
process massive amounts of information/data in parallel processing. RTAES algorithm minimizes
energy consumption in the computational, reconfiguration and communication costs; under the
scenario of parallel processing data in a cloud computing environment, while satisfying the
service level, which is expressed as the maximized time to process the job (including
communication and computational times). The main contribution of this research is; we consider
the energy objective approach, which is a non convex function. In addition, we also propose a
mathematical function to convert the non-convexity into a convex function. Moreover, our
proposed model is easy to implement, scalable and independent of workload scheduling.
The organization of the paper is as follows. Section 2 presents the existing state of art methods
related to our research in the cloud environment. Section 3 describes proposed RTAES
methodology. In Section 4 we discuss the experimental results and the performance analysis of
the proposed methodology. Finally the paper is concluded in Section 5.

3
2. RELATED WORK
The present cloud computing environment provides the necessary QoS to attract users from
different segments. In the cloud computing, several resources such as; storage, memory, CPUs
etc, where key resources are provisioned and then, they are leased to the users/clients for a
guaranteed QoS. In [18], authors have presented a novel-SLA (Service Level Agreement)
framework for cloud computing. The framework contains the reinforcement learning (RL) to
develop the hiring policy of VM, which can be considered to assure the QoS. In [19], authors
have proposed a DVFS approach that counters the trade-off between different energy
consumption level and the performance dilapidation. CPU re-allocation approach is presented in
[20] for effective energy-management in real-time services. Energy aware resource provisioning
was considered in [21]. The proposed framework considers both server and communication
network’s energy consumptions in order to provide the complete solution. Efficient energy
resources allocation model for real time services in the cloud environment is presented in[22].
There are several VMs and hosts that have been applied in order to minimize energy
consumption. The scheduling of VMs on servers to meet the energy saving parameter in the cloud
computing environment was discussed [23, 30], where a VM-scheduling approach was proposed
in order to schedule the VMs in a cloud cluster environment. Because of the higher time
complexity, this model is not as effective. Our previous work in [27], we measure the energy
consumption of a parallel application on computational grids based on processor clock cycles to
understand energy consumption issues.
Therefore, a novel scheduling approach called EVMP (Energy-aware VM-Placement) was
considered to schedule the VMs, which was more effective at reducing the power consumption.
In [8, 24], authors have proposed a meta–heuristic algorithm to minimize cost with deadline
constraints. The investigation of server power consumption with different number of VMs and
various utilizations was discussed in [9]. A VM-scheduling algorithm, which ensures the
efficient energy budget of an individual VM, was proposed in [25]. A cost constrained model
through Heterogeneous Earliest Finish-Time (HEFT) is presented in [23], in which the algorithm
provides the optimum budget in various scenarios. A similar approach is proposed in [6], where
authors proposed the heterogeneous Budget Constrained Scheduling (BCS) approach that
provided the efficient cost model by adjusting the costs and budget ratio within the assigned tasks
workflow. One multi-objective HEFT approach is a Pareto heuristic-based algorithm [7], which is
an extension to HEFT that, upgrades the costs and time factors simultaneously in order to provide
a scheduling trade-off between the objectives. The real-time scheduling of a VM in the cloud to
reduce consumption of energy was briefly addressed in [24]. Moreover, the reduction of energy
consumption can also minimize the carbon dioxide emissions (from the datacenter cooling
system) [25].
Therefore, this study mainly focuses on the VMs scheduling in order to minimize the energy
consumption in cloud environment.
3. PROPOSED RTAES ARCHITECTURE MODELING
There are several virtualized processing units that are interconnected through a single
virtual-hop network is considered in the CCVDCs model. Here the central controller is used to
manage all the devices.

4
An individual processing component executes the primary assigned task through its
computing resources and local virtual-storage. Whenever a novel task request is submitted at the
CCVDC, the performances of the location resource availability and admission control are
performed by the virtual central controller. Figure 1 shows the considered RTAES architecture,
which shows the computing and communication system model. The multiple numbers of VMs are
configured in the CCVDC where they are interconnected by a limited-switched V-LAN (virtual
local area network). Here, a star-shaped V-LAN is considered that guarantee the inter
communication of VMs and that the switching network in Fig. 1 will work as a central node. The
VM-manager generally operates both the V-LAN and VMs jointly and performs the task
scheduling activity by dynamically allocating the use of communication resources and available
virtual computing to virtual links and VMs.
Here, we consider that VMs that are deployed over the server can alter their service
resources according to the prototype presented in [31-33], this prototype has been widely adopted
in the ‘private-cloud environment’. The high computation demands in the model tend to consider
some large VMs in the place of several small VMs. In this paper, we consider a single VM that is
positioned as on individual server to simplify the model, and for that server we can use all
resources. In the physical server, the CPU functions at a frequency (𝑎𝑖) at any considered time
from the predefined existing set of frequencies that are present in the DVFS approach.
The computation of energy depends upon the VM states and the curve of the CPU energy. Here,
the DVFS is used to hosting the physical servers in order to stretch the tasks processing times
and reduce the consumption of energy by minimizing the CPU frequencies of the working VMs.
The purpose of an individual server is to work at various voltage levels and, functions at
various CPU frequencies. The set of permitted frequencies is;
𝑎(𝑖) ∈ {𝑎𝑗(𝑖)},𝑤𝑖𝑡ℎ𝑖 ∈ {1, . . . , 𝑍}, 𝑗 ∈ {0,1, . . . , 𝐵} (1)
Where, 𝐵 is the number of the ‘CPU-frequency’ that is allowed at each of the VMs, 𝑎𝑗(𝑖) is the
𝑗𝑡ℎ discrete frequency of VMi, and 𝑗 = 0represents its idle state.

5
Figure 1. Proposed RTAES architecture.
The dynamic consumption of power 𝐶 of hosting the CPU will increase with the 3rd
power of the
‘CPU frequency’. Therefore, in generic VM (𝑖) the energy consumption is
𝜕𝐹𝐶(𝑖) ≜ ∑ 𝐷𝐸𝑐𝑒𝑎𝑗(𝑖)3
𝑝𝑗(𝑖)
𝐵
𝑗=0 , [𝐽]𝑤𝑖𝑡ℎ∀𝑖 ∈ {1, . . . , 𝑍} (2)
where, 𝑝𝑗(𝑖) is the time with the 𝑗𝑡ℎ VM operator at frequency𝑎𝑗(𝑖).
The two costs are considered the switching frequency costs. In the first, the changing
costs are from the internal switching at the discrete frequencies of VM (i) from𝑎𝑗(𝑖) to 𝑎𝑗+𝑓(𝑖).In
the second, the costs are from the external switching through the final activated ‘discrete
frequency’ for the upcoming job with a job-size of𝐺𝑠𝑧. The list of discrete active frequencies for
individual VMs in each upcoming task is input into the system, and the switching activity from
the discrete active frequency to a different one disturbs the switching costs. The first discrete
active frequency is 𝑎𝑓(𝑖), and the second one is 𝑎𝑓+1(𝑖). The difference between the frequencies
is
△ 𝑎𝑓(𝑖) ≜ 𝑎𝑓+1(𝑖) − 𝑎𝑓(𝑖) (3)
The switching costs are𝐻𝑐 △ 𝑎𝑓(𝑖)2
, where the switching costs are calculate by a unit frequency
switching as 𝐻𝑐 (J/Hz2
). In the homogeneous VMs, the internal switching costs of all VMs are
𝑘𝑒 ∑ ∑ [△ 𝑎𝑓(𝑖)]2
𝐹
𝑓=0
𝑍
𝑖=1 , where 𝑓 ∈ {0, 1 , . . . , 𝐹}. 𝐹 ≤ 𝐵 is the number of discrete active
frequencies at VM (𝑖). Therefore, the total switching frequency is
∑ 𝜕𝑆𝑤(𝑖)
𝑍
𝑖=1 ≜ 𝐻𝑐 ∑ ∑ [△ 𝑎𝑓(𝑖)]2
𝐹
𝑓=0
𝑍
𝑖=1 + 𝐻𝑐 ∑ 𝐸𝑐
𝑍
𝑖=1 (4)

6
The ‘external-switching costs’ (𝐸𝑐) are considered by the multiplication of 𝐻𝑐with the last and
first discrete frequencies’ quadratic differences.
The virtual link is responsible for the communication between VM (𝑖) and the scheduler, which
operates at𝐾(𝑖) [bits/s] transmission rate (i.e., 𝑖 = 1, … , 𝑍) and is equipped with a virtual network
interface (VIF), as given in Fig. 1. The power consumed in the transmission is one-way, and the
switching function at the 𝑖 − 𝑡ℎ virtual link is
𝐿𝑐𝑜𝑛(𝑖) = 𝐿𝑐𝑜𝑛
𝑇 (𝑖) + 𝐿𝑐𝑜𝑛
𝑅 (𝑖), [𝑊] (5)
where the power consumed through the transmitted VIF is 𝐿𝑐𝑜𝑛
𝑇 (𝑖) and the receiving power with
the switching function related to VIF is𝐿𝑐𝑜𝑛
𝑅 (𝑖). Here, we consider that the receipt and
transmission are same, and it can be calculated as
𝐿𝑐𝑜𝑛(𝑖) = 𝑀𝑖(2𝑅(𝑖)/𝑁𝑖 − 1) + 𝐿𝑖𝑑𝑙𝑒(𝑖), [𝑊] (5)
where
𝑀𝑖 ≜ 𝑄0(𝑖)𝑁𝑖 /𝑟𝑖,[𝑊/𝐻𝑧] (6)
𝑟𝑖and 𝑁𝑖[Hz] are the transmission power bandwidth and the noise ‘spectral-power-density’ with
the positive gain of the ith
link. Therefore, the one-way corresponding delay in transmission is
𝑆(𝑖) = ∑
𝐴𝑗(𝑖)𝑝𝑗(𝑖)
𝐾(𝑖)
𝐵
𝑗=1 (7)
In addition, the corresponding energy at one-way communication is
𝜕𝑐𝑜𝑚(𝑖) ≜ 𝑆(𝑖)𝐿𝑐𝑜𝑚(𝑖) [𝐽𝑜𝑢𝑙𝑒] (8)
Here, our main aim is to minimize the computational energy and overall resulting
communication. This can be shown as
𝜕𝑡𝑜𝑡 ≜ ∑ 𝜕𝐹𝐶(𝑖)
𝑍
𝑗=1 + ∑ 𝜕𝑆𝑤(𝑖)
𝑍
𝑖=1 + ∑ 𝜕𝑐𝑜𝑚(𝑖)
𝑍
𝑗=1 [𝑗𝑜𝑢𝑙𝑒] (9)
where 𝜕𝑡𝑜𝑡is the overall computational energy that is formed by the different cost function of VM
(𝑖). Moreover, the problem is with the difficulty constraint (𝐸𝑡
̅̅̅) allowed in the per-job ‘execution
time’.
3.1. Proposed RTAES technique for Optimization Problem
The goal of our proposed work is to minimize the total consumption of energy (from the
incoming workload) by choosing the optimal computational resources in order to complete the
task’s execution, which depends upon the optimal bandwidth (for the consumption of energy) and
the current load level while adopting the content dependent switching frequencies for individual
VMs. The computation of resources consists of several (PMs) physical machines, where each one
comprises a single or multiple cores, a local I/O network interface and memory. The resulting

7
overall communications and energy include the 𝜕𝐹𝐶(𝑖),𝜕𝑆𝑤(𝑖) and 𝜕𝑐𝑜𝑚(𝑖) for 𝑍 number of VMs
in the 𝜕𝑡𝑜𝑡, which are estimated for individual VMs.
To optimize the problem, it is important to apply an additional parameter 𝐸𝑡, which is the
threshold value in the task’s processing operation. Therefore, the service level is split into two
constraints, which are that the network-dependent time is less than ‘or’ equal to 𝐸𝑡
´ − 𝐸𝑡 and that
the computational time is less than or equal to 𝐸𝑡. Therefore, the resulting equation of the
optimization problem can be expressed as;
(10)
Equation (10) consists of three major terms to find the computational energy. The decision
variable for optimizing the difficulty is 𝑝𝑗(𝑖), where the time duration is 𝑝𝑗(𝑖) for each VM
operating at frequency 𝑎𝑗frequency and the transmission rate for each VM (𝑖) is 𝐾(𝑖).
𝐺𝑠𝑧 = ∑ ∑ 𝐴𝑗(𝑖)𝑝𝑗(𝑖)
𝐵
𝑗=0
𝑍
𝑗=1
(11)
The above equation (11) is a global constraint that insures that the overall task is split
into 𝑍number of parallel tasks. The 𝐺𝑠𝑧 is incoming job size, which needs to spread over 𝑍
number of VMs during computation.
(12)
The inequality bandwidth in the above equation (12) ensures that the summation of the bandwidth
at each VM should be less than the maximum global network’s bandwidth.
∑ 𝑝𝑗(𝑖)
𝐵
𝑗=0
≤ 𝐸𝑡, 𝑖 = 1, … , 𝑍
(13)
This is the computational time constraint 𝑇.
∑
2𝐴𝑗(𝑖)𝑝𝑗(𝑖)
𝐾(𝑖)
𝐵
𝑗=0
≤ 𝐸𝑡
̅̅̅ − 𝐸𝑡, 𝑖 = 1, … , 𝑍
(14)
Equation (14) describes the constraint at the data exchange period/time.
0 ≤ 𝑝𝑗(𝑖) ≤ 𝐸𝑡,𝑖 = 1, … , 𝑍, 𝑗 ∈ {0, . . . , 𝐵} (15)
0 ≤ 𝐾(𝑖) ≤ 𝐾𝑡 , ∀𝑖 ∈ {1 , . . . , 𝑍} (16)
𝑀𝑖𝑛𝑖𝑚𝑖𝑧𝑒 ∑ 𝜕𝐹𝐶(𝑖)
𝑍
𝑗=1 + ∑ 𝜕𝑆𝑤(𝑖)
𝑍
𝑖=1 + 𝜕𝑐𝑜𝑚(𝑖)
∑ 𝐾(𝑖)
𝑍
𝑗=1
≤ 𝐾𝑡
(12)

8
Equation (15) guarantees the time of individual computing intervals (i.e., positive and less
than𝐸𝑡). The final constraint ensures that the 𝐾 control parameter (i.e., the channel
communication rate) should be positive and less than the maximum network capacity. The power-
cost at the end-to-end link comes from equation (8) and the outcome can be given as
∑
2𝐿𝑐𝑜𝑚(𝑖)𝐴𝑗(𝑖)𝑝𝑗(𝑖)
𝐾(𝑖)
𝐵
𝑗=0
(17)
Where 𝐾(𝑖) > 0 and∀𝑖 ∈ {1 , . . . , 𝑍}. withThe multi-variable coefficient value and arrow sign
implies that
∑
2 𝐴𝑗(𝑖)𝑝𝑗(𝑖)
𝐾(𝑖)
≤ 𝐸𝑡
̅̅̅ − 𝐸𝑡 → (∑
𝐴𝑗(𝑖)𝑝𝑗(𝑖)
𝐾(𝑖)
𝐵
𝑗=0
)
𝐵
𝑗=0
≤
𝐸𝑡
̅̅̅ − 𝐸𝑡
2
(18)
The above equation (18) is obtained from equation (13). Furthermore, to make it easier to
optimize the problem, here we modify the ‘second control’ variable through 𝐾(𝑖)with the other
control variable𝑝𝑗(𝑖)as
∑
𝐾(𝑖)
≤ 𝐸𝑡
̅̅̅ − 𝐸𝑡 → 𝐾(𝑖) ≥ ∑
𝐸𝑡
̅̅̅ − 𝐸𝑡
𝐵
𝑗=0
𝐵
𝑗=0
(19)
The outcome of equations (18) and (19) is applied to get the 𝜕𝑐𝑜𝑚(𝑖) end-to-end function that is
dependent upon the control variable 𝑅 (𝑈(𝑖);𝑝𝑗(𝑖)), which can be modified by altering the
control variable 𝑈(𝑖)to the other ‘control variable’𝑝𝑗(𝑖) in eq. (20).
𝜕𝑐𝑜𝑚(𝑖) = 𝑅 (𝑈(𝑖);𝑝𝑗(𝑖)) ≜ 𝑉 (𝑎𝑗(𝑖)) (20)
Eq. (20) represents the formula for the adaptive energy communication in the end-to-end link,
which depends upon the time variables’ simulation for each VM and the main function 𝑉(. ).that
satisfies the third term in eq. (12) is convex.
4. RESULT AND ANALYSIS
Over the past several years, the embedded processor demand has become extremely high
worldwide due to the vast usage of network equipment, information devices, portable gadgets and
digital instruments. Supreme performance is always required from the embedded equipment
because it is used in our daily lives, such as in the multimedia ‘digital-signal-processing’
technique. However, these devices that are equipped with embedded processors have some
drawbacks that affect their device’s efficiency due to the consumption of high power by the
devices. Moreover, there are performance imbalances between devices. Therefore, here we
evaluate the performance of our proposed model and existing models with respect to power
consumption. Several results are presented using our proposed RTAES algorithm that is based
upon the DVFS approach. In this section, we have considered the different jobs of ‘30’,
‘50’,‘100’ and ‘1000’ in order evaluate the execution time and the ‘𝑀𝑜𝑛𝑡𝑎𝑔𝑒’ scientific dataset

9
(MSD) has been adapted to validate our model. The power consumption and execution time can
be evaluated using several parameters, which are shown in below Table 1. This model is
simulated using cloudsim and implemented on the Windows 10, 64-bit operating system and uses
a computer with an i5 processor with 8 GB RAM and the programming language used for code is
JAVA.
4.1. Comparative Analysis
In the present technical era, cloud computing has conquered the world market in several areas,
such as trading, healthcare, the software sector and medical. Thus, the upcoming technologies are
largely based on cloud computing processor operations because of its extensive demand. The lack
of accurate resource utilization can affect their device efficiency due to their high power
consumption; therefore, to address these issues, here we introduce the RTAES algorithm that is
based upon the DVFS approach. The effective scheduling of tasks can improve users’
interactions, avoid task overloads, and enhance resource utilization and throughput. Therefore,
here, we presented the comparison of results with existing techniques with respect to the
consumption of power and operational execution time.
Table 1: Various parameters comparison for proposed technique vs DVFS using MSDX
Parameters
Total
execution
time (sec)
Energy
consumption
(𝑊ℎ)
Power
Sum
(W)
Power
Average
(W)
DVFS
MSD-25
VM-
100
174.67 482.29 1663184 95.22
MSD-50 VM-80 387.53 944.14 2953095 76.2
MSD-100 VM-60 817.35 1839.12 4673928 57.18
MSD-
1000
VM-40 8591.39 70377.1 3.3E+07 38.16
RTAES
MSD-25
VM-
100
57.72 120 417907 72.4
MSD-50 VM-80 71.77 124.88 418104 58.25
MSD-100 VM-60 101.91 143.79 449547 44.11
MSD-
1000
VM-40 777.04 1046.14 2328420 29.96
The several parameters that are used in Table 2 are the total simulation time, the energy
consumption, the power sum and the power average. The scientific model of Montage was
considered with different job sizes, such as MSD-25, MSD-50, MSD-100 and MSD-1000.
Moreover, each MSD that we have considered was tested with different virtual machines (VMs)
that are shown in Table1 above. The energy consumption values are 482.29Wh for MSD-25,
944.14Wh for MSD-50, 1839.12Wh for MSD-100 and 70377.1Wh for MSD-1000. As shown in
Table 1, these values are substantially lower than those of the existing technique. Here, we
obtained 75%, 86%, 92% and 98% less energy consumption with respect to the DVFS approach.

10
The total simulation times in the existing approach for different MSDs are 174.67, 387.53, 817.35
and 8591.39 sec, which consumed 66%, 81%, 87% and 90% more time compared to that of our
proposed model respectively. Table 2 represents the Average Execution time (sec) comparison of
the proposed technique with other existing techniques using MSD. The proposed RTAES
approach calculates the average execution times for MSD-25 as 2.3 sec, MSD-50 as 1.4 sec,
MSD-100 as 1.02 sec and MSD-1000 is 0.77 sec, which are substantially lower than those with
existing techniques such as EMOA [26] and DVFS.
Table 1: Average execution time (sec) comparison of the proposed technique with other existing techniques using MSD
Datasets Number of
nodes
Average Execution time (s)
EMOA [34] DVFS RTAES
MSD-25 30 8.44 6.9868 2.3088
MSD-50 50 9.78 7.7506 1.4354
MSD-100 100 10.58 8.1735 1.0191
MSD-1000 1000 11.36 8.59139 0.77704
4.2. Graphical Representation
Here, we represent the graphical form of our obtained results. Figure 2 shows the total simulation
time comparison of the existing DVFS model and the proposed RTAES approach using the
Montage scientific model datasets for different numbers of nodes and VM numbers. Figure 3
shows the energy consumption comparison of the existing DVFS model and the proposed
RTAES approach using the Montage scientific model datasets for different numbers of nodes and
VM numbers. Figure 4 shows the power sum comparison of the existing DVFS model and the
proposed RTAES approach using the Montage scientific model datasets for different numbers of
nodes and VM numbers. Figure 5 shows the power average comparison of the existing DVFS
model and the proposed RTAES approach using the Montage scientific model datasets for
different numbers of nodes and VM numbers. Figure 6 shows the average execution time (s)
comparison of the existing DVFS model and the proposed RTAES approach using the Montage
scientific model datasets for different numbers of nodes and VM numbers.
Figure 2: Total simulation time comparison of the existing and the proposed model using MSD
174.67 387.53 817.35
8591.39
57.72 71.77 101.91
777.04
0
2000
4000
6000
8000
10000
VM-100 VM-80 VM-60 VM-40
MSD-25 MSD-50 MSD-100 MSD-1000
DVFS RTAES

11
Figure 3: Energy consumption comparison of the existing and the proposed model using MSD
Figure 4: Power comparison of the existing and the proposed model using MSD
Figure 5: Power average comparison of the existing and the proposed model using MSD
482.29 944.14 1839.12
70377.06
120 124.88 143.79 1046.14
0
20000
40000
60000
80000
VM-100 VM-80 VM-60 VM-40
DVFS RTAES
1663184.16 2953094.75 4673928.26
32789145.44
417907.46 418103.69 449547.24 2328420.37
0
5000000
10000000
15000000
20000000
25000000
30000000
35000000
VM-100 VM-80 VM-60 VM-40
DVFS RTAES
95.22
76.2
57.18
38.16
72.4
58.25
44.11
29.96
0
20
40
60
80
100
VM-100 VM-80 VM-60 VM-40
DVFS RTAES

12
Figure 6: Average execution time (sec) comparison of the existing techniques and the proposed model
using MSD
5. CONCLUSION
Task scheduling in embedded processors to minimize the consumption of energy in individual
VMs, has essential significance. Therefore, in this paper, we proposed the real time adaptive
energy-scheduling (RTAES) algorithm by manipulating the reconfiguration proficiency of cloud
computing-virtualized data centers (CCVDCs) by processing massive amounts of information /
data of an application. The RTAES algorithm minimizes the consumption of energy in the
computational, reconfiguration and communication costs under the scenario of parallel processing
data in a cloud computing environment. The RTAES algorithm is based on DVFS. We also tried
to minimize three cost functions, including communication costs, computational costs, and
switching costs, to improve the system’s performance. The modeling details were presented
above. The Montage scientific dataset (MSD) was used to validate our model with respect to
other existing techniques. The above results were demonstrated in terms of the total execution
time, the decrease in power consumption and the average power that was essential for processor
operation. Using our proposed RTAES model, the total simulation times are 57.72 sec for MSD-
25, 71.77 sec for MSD-50, 101.91 sec for MSD-100 and 777.04 sec for MSD-1000, values that
are substantially lower than those of the existing DVFS technique. Similarly, the energy
consumption values using our proposed RTAES model are 120Wh for MSD-25, 124.88Wh for
MSD-50, 143.79Wh for MSD-100 and 1046.14Wh for MSD-1000, values that are substantially
lower than those of the existing DVFS technique. Therefore, the results of our proposed RTAES
model demonstrate its effectiveness in power consumption, performance and execution time with
respect to other existing approaches.
REFERENCES
[1] Lizhe Wang, Gregor von Laszewski, Jai Dayal, Thomas R. Furlani, Thermal aware workload
scheduling with backfilling for green data centers, in: IPCCC, 2009, pp. 289–296.
[2] William Forrest, How to cut data centre carbon emissions? Website, December 2008.
8.44
9.78
10.58
11.36
6.9868
7.7506 8.1735 8.59139
2.3088
1.4354 1.0191 0.77704
0
2
4
6
8
10
12
Montage-25 Montage-50 Montage-100 Montage-1000
EMOA [34] DVFS RTAES

13
[3] Yongpan Liu et al. “Thermal vs energy optimization for dvfs-enabled processors in embedded
systems”. In: Quality Electronic Design, 2007. ISQED’07. 8th International Symposium on. IEEE.
2007, pp. 204–209.
[4] J.-J. Chen and C.-F. Kuo, `Ènergy-efficient scheduling for real-time systems on dynamic voltage
scaling (DVS) platforms,'' in Proc. IEEE RTCSA, Aug. 2007, pp. 28-38.
[5] V. Devadas and H. Aydin, ``Coordinated power management of periodic real-time tasks on chip
multiprocessors,'' in Proc. Green Comput. Conf., 2010, pp. 61-72.
[6] Arabnejad, H., Barbosa, J.G.: A budget constrained scheduling algorithm for workflow applications.
J. Grid Comput. 12(4), 15 (2014)
[7] Durillo, J.J., Prodan, R.: Multi-objective workflow scheduling in amazon ec2. Cluster Comput. 17(2),
169–189 (2014)
[8] Q. Wu, F. Ishikawa, Q. Zhu, Y. Xia and J. Wen, "Deadline-Constrained Cost Optimization
Approaches for Workflow Scheduling in Clouds," in IEEE Transactions on Parallel and Distributed
Systems, vol. 28, no. 12, pp. 3401-3412, Dec. 2017.
[9] S. Chinprasertsuk and S. Gertphol, "Power model for virtual machine in cloud computing," 2014 11th
International Joint Conference on Computer Science and Software Engineering (JCSSE), Chon Buri,
2014, pp. 140-145.
[10] T. AlEnawy et al., `Ènergy-aware task allocation for rate monotonic scheduling,'' in Proc. IEEE
RTAS, Apr. 2005, pp. 213-223.
[11] C.-Y. Yang, J.-J. Chen, T.-W. Kuo, and L. Thiele, `Àn approximation scheme for energy- efficient
scheduling of real-time tasks in heterogeneous multiprocessor systems,'' in Proc. DATE, 2009, pp.
694-699.
[12] E. Seo, J. Jeong, S. Park, and J. Lee, `Ènergy efcient scheduling of realtime tasks on multicore
processors,'' IEEE Trans. Parallel Distrib. Syst., vol. 19, no. 11, pp. 1540-1552, Nov. 2008.
[13] C. Xian and Y.-H. Lu, ``Dynamic voltage scaling for multitasking real-time systems with uncertain
execution time,'' in Proc. ACM GLSVLSI, 2006, pp. 392-397.
[14] C. Xian, Y.-H. Lu, and Z. Li, `Ènergy-aware scheduling for real-time multiprocessor systems with
uncertain task execution time,'' in Proc. ACM DAC, 2007, pp. 664-669.
[15] Baliga, J, Ayre, R. W., Hinton, K., and Tucker, R. 2011. Green cloud computing: Balancing energy in
processing, storage, and transport. Proceedings of the IEEE, 2011. 99(1):149–167.
[16] Canali, C. and Lancellotti, R. Parameter Tuning for Scalable Multi-Resource Server Consolidation in
Cloud Systems. Communications Software and Systems, 2016.
[17] Urgaonkar, R., Kozat, U. C., Igarashi, K., and Neely, M. J. Dynamic resource allocation and power
management in virtualized data centers. In NOMS, 2010, pp 479–486. IEEE.
[18] A. Alsarhan, A. Itradat, A. Y. Al-Dubai, A. Y. Zomaya and G. Min, "Adaptive Resource Allocation
and Provisioning in Multi-Service Cloud Environments," in IEEE Transactions on Parallel and
Distributed Systems, vol. 29, no. 1, pp. 31-42, Jan. 1 2018.

14
[19] P. Arroba, J. M. Moya, J. L. Ayala and R. Buyya, "DVFS-Aware Consolidation for Energy-Efficient
Clouds," 2015 International Conference on Parallel Architecture and Compilation (PACT), San
Francisco, CA, 2015, pp. 494-495.
[20] W. Chawarut and L. Woraphon, "Energy-aware and real-time service management in cloud
computing," 2013 10th International Conference on Electrical Engineering/Electronics, Computer,
Telecommunications and Information Technology, Krabi, 2013, pp. 1-5.
[21] H. Faragardi, A. Rajabi, K. Sandström and T. Nolte, "EAICA: An energy-aware resource
provisioning algorithm for Real-Time Cloud services," 2016 IEEE 21st International Conference on
Emerging Technologies and Factory Automation (ETFA), Berlin, 2016, pp.1-10.
[22] Z. Bagheri and K. Zamanifar, "Enhancing energy efficiency in resource allocation for real-time cloud
services," Telecommunications (IST), 2014 7th International Symposium on, Tehran, 2014, pp. 701-
706.
[23] Zheng, W., Sakellariou, R.: Budget-deadline constrained workflow planning for admission control. J.
Grid Comput. 11(4), 105–119, 2013.
[24] T. Knauth and C. Fetzer, “Energy-aware scheduling for infrastructure clouds,” in Cloud Computing
Technology and Science (CloudCom), 2012,IEEE 4th International Conference on. IEEE, 2012, pp.
58–65.
[25] S. K. Garg, C. S. Yeo, and R. Buyya, “Green cloud framework for improving carbon efficiency of
clouds,” in Euro-Par 2011 Parallel Processing. Springer, 2011, pp. 491–502.
[26] Z. Zhu, G. Zhang, M. Li and X. Liu, "Evolutionary Multi-Objective Workflow Scheduling in Cloud,"
in IEEE Transactions on Parallel and Distributed Systems, vol. 27, no. 5, pp. 1344-1357, May 1 2016.
[27] D.B Srinivas, Puneeth R.P, Rajan M.A, Sanjay. H.A, An integrated framework to measure the
energy consumption of a parallel application on a computational grid, International Journal of
Computer Science and Network Security, VOL.16 No.7, pp. 94-98, 2016,

REAL-TIME ADAPTIVE ENERGY-SCHEDULING ALGORITHM FOR VIRTUALIZED CLOUD COMPUTING

Recommended

More Related Content

Similar to REAL-TIME ADAPTIVE ENERGY-SCHEDULING ALGORITHM FOR VIRTUALIZED CLOUD COMPUTING (20)

Recently uploaded (20)

REAL-TIME ADAPTIVE ENERGY-SCHEDULING ALGORITHM FOR VIRTUALIZED CLOUD COMPUTING