Energy Efficiency Policies for Smart Digital Cloud Environment

based on Heuristics Algorithms

Awatif Ragmani, Amina El Omri, Noreddine Abhgour, Khalid Moussaid and Mohamed Rida

Department of Mathematics and Computer Science, Faculty of Sciences, University Hassan II, Casablanca, Morocco

Keywords: Cloud Computing, Energy Efficiency, Scheduling, Load Balancing, Metaheuristics, Greencloud.

Abstract: The Cloud computing model is based on the use of virtual resources and their placement on physical servers

hosted in the different data centers. Those data centers are known to be big energy consumers. The allocation

of virtual machines within servers has a paramount role in optimizing energy consumption of the underlying

infrastructure in order to satisfy the environmental and economic constraints. Since then, various hardware

and software solutions have emerged. Among these strategies, we highlight the optimization of virtual

machine scheduling in order to improve the quality of service and the energy efficiency. Through this paper,

we propose firstly, to study energy consumption in the Cloud environment based on the GreenCloud

simulator. Secondly, we define a scheduling solution aimed at reducing energy consumption via a better

resource allocation strategy by privileging data center powered by clean energy. The main contributions of

this paper are the use of the Taguchi concept to evaluate the Cloud model and the introduction of scheduling

policy based on the simulated annealing algorithm.

1 INTRODUCTION

The last decade has perceived an important expansion

of Cloud computing due to its practical and economic

aspect. However, this growth goes with a tremendous

increase in energy consumption. Indeed, Cloud

computing services require huge data center that

consumes energy in order to provide the necessary

elasticity and scalability to their customers. In 2011,

Google has announced that its energy consumption

was around 2,675,898 MWh (Aschberger and

Halbrainer, 2013). More globally, the quantity of

electricity used by data centers has been estimated at

around 1.5% of the world's total electricity

consumption. As depicted in Figure 1, only 15% of

the energy utilized by a data center is used for

computing purposes which open a wide scope for

possible energy efficiency solutions. In addition, only

40% of the energy is distributed to IT equipment

which includes computer servers and communication

equipment; while 60% of the energy is distributed

between the cooling system and the system of energy

distribution. In summary, the Gartner team has

identified that energy expenditure is equivalent to

almost 10% of current data center operational

expenditure (OPEX) and tends to increase over the

next decade to nearly 50% (Kliazovich et al., 2013a).

Thus, it is essential to rationalize the energy

consumption while guaranteeing the highest level of

quality of service for the customer. Based on the

previous findings, it becomes clear that alternative

solutions must be found in order to curb this

exponential growth. Currently, the two most applied

concepts for energy saving in computer systems are

Dynamic Power Management (DPM) and Dynamic

Voltage and Frequency Scaling (DVFS). The first

concept is based on a dynamic management of

frequency and voltages according to the level of

performance required. The second concept allows us

to turn off or put on standby the inactive servers.

However, both of these concepts have some

shortcomings in terms of performance. Through this

article, we introduce a new strategy to analyze and

optimize the energy efficiency within a data center

based on Taguchi concept and metaheuristics

algorithms. The main motivation behind this article is

the evaluation of the parameters that have the most

influence on energy efficiency in order to propose

smart policies able to reduce energy losses in a data

center.

In summary, the problem of energy efficiency in a

data center can be dealt with at two levels. The first

354

Ragmani, A., El Omri, A., Abghour, N., Moussaid, K. and Rida, M.

Energy Efﬁciency Policies for Smart Digital Cloud Environment based on Heuristics Algorithms.

DOI: 10.5220/0006671403540363

In Proceedings of the 8th International Conference on Cloud Computing and Services Science (CLOSER 2018), pages 354-363

ISBN: 978-989-758-295-0

Figure 1: The distribution of energy consumption.

level is the quality of the hardware used in the data

center, notably the category of servers and the

typology of the network. The second level includes

all data management procedures including

scheduling, load balancing, virtual machine

migration, fault tolerance, and security strategies. The

optics of concentrating the virtual machines on a

smaller number of servers is motivated by the fact that

an idle server consumes two-thirds of its maximum

consumption and the last tier fluctuates linearly with

the processes in process of treatment (Zhou et al.,

2015). In other words, an efficient planning strategy

ensures the consolidation of virtual machines within

a smaller number of physical servers and enables

power to be removed from the rest of the servers

(Sosinsky, 2011; Velte et al., 2010). The rest of paper

is organized as follows: Section 2, summarizes the

research works which have tackled the issues of

scheduling and energy efficiency within a distributed

computing system. Section 3 introduces the state of

art and background to evaluate the energy efficiency

of Cloud environment. Section 4 describes the

problem statement, the methodology applied to

evaluate the energy within the Cloud and, the

proposed scheduling strategies for virtual machines

allocation. Finally, conclusions are strained in

Section 5.

2 RELATED WORK

In recent years, the energy efficiency of a data center

has been of major importance. Several research

studies have addressed this issue from diverse points

of view in order to propose different solutions to

enhance the energy efficiency of Cloud data center.

Particularly, the authors of (Chinnici and Quintiliani,

2013) introduced an evaluation methodology of

energy saving which has been carried out following

the replacement of a set of components by more

energy efficiency mechanisms. The concept proposed

by the authors was put in place to match a

comprehensive strategy to encourage the adoption of

energy efficiency policies in developing countries. In

addition, the authors (Zhou et al., 2015) proposed a

solution called (TESA) for deploying virtual

machines to optimize the energy efficiency of a large-

scale data center. This solution is based on the linear

relationship between energy consumption and

resources use. The TESA solution suggests migrating

virtual machines hosted by hyper-loaded machines to

less-loaded machines. In addition, the authors

introduced five types of virtual machine selection

policies. Another study (Marotta and Avallone, 2015)

pointed out the positive impact of the application of

strategies based on the consolidation of virtual

machines on the waste of energy in data centers. The

proposed solution is based on the principles of

migrating virtual machines in order to reduce the

number of active servers while guaranteeing the best

cost-quality compromise. The main contribution of

this study lies in the introduction of a machine

consolidation model inspired by the simulated

annealing algorithm. The proposed algorithm is based

on calculating the attractiveness value of virtual

machine migrations by having several input

parameters. The defined solution allows outlining the

list of virtual machines migration by reducing the

power consumption of the active servers. The

research paper (Guzek et al., 2013) defines a holistic

model of a data center of the Cloud. The introduced

model describes Cloud applications, physical

machines, and virtual machines through the

description of parameters such as memory and

storage. The model was tested within the well-known

GreenCloud simulator. The authors announce that the

proposed model improves the accuracy of the

simulations. The authors also plan to develop in the

future a new module for managing the migration of

virtual machines. The authors of (Sarji et al., 2011)

tackle the static analysis of a server before proposing

two models of energy management in order to ensure

the energy efficiency in the Cloud computing

environment. The operating principle of these models

allows analyzing the state of the system before

deciding on the migration of the virtual machines

from one physical node to another. Then, the

proposed system turns the idle servers off or on

standby. The two strategies proposed by the authors

are differentiated by the energy consumed and the

start-up time of a server previously put into standby

or off mode. Thus, the authors leave the choice to the

system administrator to select the level of energy

efficiency or the response time required by the SLA.

In the research study (Beloglazov et al., 2012), the

authors proposed to use heuristic solutions to

Energy Efﬁciency Policies for Smart Digital Cloud Environment based on Heuristics Algorithms

355

guarantee energetic efficiency combined with a

suitable level of quality of service. This study

includes an analysis of existing solutions and

possibilities for optimizing energy while taking into

account quality of service constraints. The authors of

the research paper (Aschberger and Halbrainer,

2013), announce that the processing of a task or the

processing of a virtual machine has different impacts

on the consumption of energy within the Cloud

environment. They stated that task management is

more compatible with energy saving because many

parameters are known in advance such as CPU usage

and execution time. This last parameter allows

predicting the moment when a resource becomes free

again. On the other hand, the authors indicate that

virtual machines could be allocated to physical

machines without specifying the end date of the

virtual machine which could continue to exist as long

as the client pays the created instance. Finally, the

authors of the research paper (Canali et al., 2017)

stressed the complexity of the minimization of energy

consumption within the data center and the aspect of

separating the solutions used from the IT and

communication aspects. The authors specify that this

separation has a negative impact on energy efficiency

in the data center. As a result, they introduce a

powerful model that takes into account both aspects

of machine allocation management, which improves

energy efficiency. The proposed model embraces

three advantages, including the consideration of data

traffic exchanges between virtual machines, the

modeling of the energy consumption relating to the

virtual machine migration, and finally the

consideration of several weighting parameters in the

process of allocating virtual machines. The authors

confirm that the proposed schema has outperformed

previous approaches of virtual machine allocation in

terms of energy efficiency. In summary, we

emphasize the disparity of approaches used by

researchers to reduce the energy consumption of data

centers, which raises the problem of finding the most

efficient solution and the possibility of combining

several solutions for an even better energy efficiency

while respecting the quality of service requirements.

3 PRELIMINARIES

3.1 Virtual Machine Migration

The virtualization has opened diverse possibilities to

share memory and processor resources in order to

create virtual machines that can support query

processing. Nowadays, virtualization is widely

applied in Cloud data centers for the purpose of

directing physical resources through partitioning,

consolidation, and isolation. Primarily, virtual

machine migration is used to allocate and reassign

resources by moving an application from one server

to another server with better technical characteristics

such as power, memory, or power consumption (Jin

et al., 2011; Marotta et al., 2018). The virtual machine

migration policy includes the procedure for triggering

the virtual machine migration process, the

identification procedure for the target virtual

machine, and the procedure of selecting a destination.

First, the triggering procedure specifies the nature of

the migration functions. The target virtual machine

identification procedure allows the selection of

machines that need to migrate for security or

performance reasons. Finally, the destination

selection procedure is applied to identify the

destination servers that will host the virtual machines

to migrate (Li et al., 2013). Live virtual machine

migration is the set of procedures for moving a

functional virtual machine from one host server to

another using the applied virtualization solution.

Finally, the migration improves load balancing, IT

maintenance and fault tolerance (Sallam and Li,

2014).

3.2 GreenCloud Simulator

GreenCloud is a Cloud simulation platform focused

on the various technical aspects of data centers and

which has proved its worth in recent years, notably in

terms of energy efficiency evaluation. This

simulation platform has been developed on the basis

of the ns-2 simulator. It allows configuring the

typology of the network, the energetic configurations

as well as other parameters of the Cloud environment

Through its graphical interface, it is possible to define

an infinite number of scenarios (Kliazovich et al.,

2012). Following each simulation GreenCloud

generates a detailed report of the energy consumption

by kind of component. Particularly, GreenCloud

offers the possibility of configuring several

scheduling strategies in order to compare their

performance. The main strength of GreenCloud

remains its flexibility to complete detailed analyzes

of the workload within the data center while insisting

on the simulations of packet communications. The

simulator presents four network topologies already

defined. In addition, GreenCloud offers the

possibility to define custom topologies. The

GreenCloud simulator takes into account several

three-tier data center architectures such as DPillar,

DCell, BCube, and FiConn. This three-tier

CLOSER 2018 - 8th International Conference on Cloud Computing and Services Science

356

architecture (see Figure 2) includes the core level, the

aggregation level that handles routing, and the access

level which manages servers in racks Each rack can

take up to 48 servers. Note that interconnections

between the servers are achieved via 1 Gigabit

Ethernet links. The simulator assumes that an inactive

server consumes nearly two-thirds of its peak load,

and the remaining tier varies with the workload of the

server. Moreover, the power consumption of the

network switches is matched to the port transmission

rate, the type of switch, number of ports, and cabling

solutions used (Guzek et al., 2013).

4 PROPOSED SOLUTION

4.1 Problem Statement

Through this article, we strive to analyze the

optimization of energy efficiency in the Cloud. Based

on the benchmarks and conclusions of the earlier

research papers summarized in the related works

section, it emerges that the most effective approach to

ensure an energy saving policy is the minimization of

the number of the running servers. However, reducing

the number of physical machines cannot be done

without impacting the performance of Cloud

applications. In other words, the techniques currently

applied reduce the energy consumed but increase the

response time of the system. However, Cloud

customers expect to have a responsiveness of services

as close as when using local services. In addition, the

principle of elasticity of the Cloud requires that

resource planning policies should be flexible, both

upward and downward. It is therefore essential to

define intelligent algorithms capable of managing the

creation and removal of virtual machines in such a

way as to respect both the energy saving constraint

and the response time constraint. In addition,

considering the environmental impact of a data

center, we propose to introduce a parameter relating

to the nature of the energy used in order to give an

advantage to a data center which uses renewable

energies. The evaluation of the energy consumed in

the data center is carried out on the basis of several

key performance indicators. Particularly, the Green

Grid consortium has established two indicators which

are power utilization efficiency (PUE) and data center

infrastructure efficiency (DCiE), summarized below:

=







(1)

Figure 2: GreenCloud three-tier architecture.

=



=







×100

(2)

 and  are calculated based on the ratio of

the total amount of energy consumed by the data

center to the amount of energy consumed by the IT

equipment. The  indicator is the inverse of the

 and is written as a percentage. Thus, a 

value close to 100% means that the data center does

not waste a lot of energy and can be defined as

efficient. In brief, the improvement of energy

efficiency in the Cloud environment is achieved by

reducing the sum of the energy consumed by the data

center components at different activity levels

(standby, full load, power off ...). Based on the

various studies considered, the components of the

data center has been classified into three categories:

IT equipment which includes servers and network

equipment, cooling equipment and power distribution

equipment (Aschberger and Halbrainer, 2013). The

evaluation of energy efficiency in data centers first

requires the identification of the elements that

consume electrical energy according to the different

levels of workload. As depicted in Figure 1 and

according to the formula (3), the energy consumed is

a function of time and can be represented as an

integral part of the energy consumption function over

a period of time. Thus, the energy consumption of a

data center (DC) which has  nodes, between an

instant 1 and 2 is calculated according to the

equations below:

=()×

(3)





=(







+





)−



(4)





=(P



+





)×∆−E





(5)

Energy Efﬁciency Policies for Smart Digital Cloud Environment based on Heuristics Algorithms

357

Where 



is the average power of the IT equipment,





is the average power of the cooling system, 



is the average power of distributing system, and





is the energy produced by whole data center

components such as the heat produced by the servers

and reused in the air conditioning systems or to warm

water.





=



+



+



(6)





= (





+









+





+





)

(7)

Where 



is the energy consumption of the 

nodes of the data center, 



is the power

consumed by the whole switching components, and





includes the energy consumption of the

auxiliary parts as well as the energy loss. According

to (Chinnici and Quintiliani, 2013), the energy

consumed related to CPU depends on the load of the

node and could be expressed as below:







=







(



)

(

ℎ





)

×

(8)

In summary, all studies agree that suspending and

activating servers according to the workload in the

data center remains the most efficient method of

ensuring overall energy efficiency including

equipment which ensures the proper functioning of a

data center. Nevertheless, continuous modification in

power modes causes significant delays. In addition,

the solution must take into account the case of the

inability to wake up a dormant server by quickly

making available another inactive server that can

receive a request without impacting the quality of

service. Further research works (Sarji et al., 2011)

have confirmed the importance of the operating

system (OS) layer in the process of setting up inactive

devices and thereby improving energy efficiency.

Lastly, the use of the technique (DVFS) to change the

server voltage in order to decrease the energy

consumed without violating SLA requirements is also

a technique that has proved its worth. However,

having a narrow number of statuses that can be

scheduled on the basis of the frequency and voltage

of the data combined to the fact that it is not adapted

to the other elements of the data center limits the gain

that can be derived from the DVFS technique. Thus,

we aim through this article to evaluate the energy

efficiency by applying the concept of Taguchi.

4.2 Problem Analysis Methodology

In order to evaluate the main factors which could

impact the energy efficiency of a data center, we have

used the GreenCloud simulation platform and the

performance analysis methodology which has been

defined in a previous paper (Ragmani et al., 2016a).

The main idea of the proposed methodology is to

describe a complex system by means of a set of inputs

that represent the most influential factors and outputs

that translate the key performance indicators (see

Figure 3) instead of evaluating all the system’s

components. Then, the Taguchi experiment plans are

used to study complex technical problems by

analyzing the various parameters that could influence

the effectiveness of the system. The performance to

analyze is represented by one or more responses such

as the response time and the energy consumed.

Experiment plans make it possible to evaluate the

parameters responsible for the variations of each

response. In short, the Taguchi experiment plan is a

set of trials arranged in advance so as to identify in a

minimum of manipulations and with a maximum of

precision the influence of multiple parameters on one

or more responses. The success of the performance

analysis approach according to the Taguchi concept

depends on the respect of the following steps:

 Formalize the problem to study by defining the

influential factors and key performance;

 Select the parameters, define their variation

levels and select their interactions;

 Build the experiment plan according to the

Taguchi tables;

 Carry out the tests;

 Analyze the results;

 Conclude after choosing the setting parameters

that can be controlled and achieve the

confirmatory test.

The applied Taguchi plans depend on the number

of modalities per parameter as well as the number of

interactions (Taguchi et al., 2005). Parameters are

assigned to columns taking into account interactions

and parameters that are difficult to modify. In our

case, we have 19 factors including 7 factors with two

levels and 12 factors with 3 levels. These parameters

correspond to the simulation variables defined by the

GreenCloud simulator and which make it possible to

describe the simulation scenarios. But the simulator

does not identify which parameter has the most

influence on the energy efficiency and also what

value must each parameter take in order to have the

best energy consumption. For these reasons, we have

performed this analysis. The description of the factors

CLOSER 2018 - 8th International Conference on Cloud Computing and Services Science

358

and their level is presented in Table 1. According to

the Taguchi tables predefined in the Minitab toolkit

and based on the number of factors, we have applied

the L36 matrix which corresponds to 36 trials. The

results are analyzed according to two complementary

modes. On the one hand, the graphical analysis that

allows representing the influence of the parameters

and their interactions. On the other hand, the

statistical analysis of the variance aims to separate, in

the global variations of the answer, the part due to the

real influence of the parameters of the part due to

chance. In brief, each factor of the inputs can take

several values identified by levels (see Table 2).

These values may be quantitative or qualitative (see

Table 1). The Taguchi method relies on the

calculation of the signal-to-noise ratio (SNR) in order

to rank factors based on their influence. According to

this concept, the optimization of outputs could be

achieved by minimizing the function described in the

equation (9).

SNR



=−10log















(9)

Where : experiment number; : trial number; Ni:

Number of trials for experiment and 



: performance

representative measurements per trial.

The performance evaluation cannot be done

without key indicators that ensure an objective

assessment of the various technical and economic

aspects of a system. Thus, we have selected five key

performance indicators that include the switches

energy consumed by the core level, the switches

energy consumed by the aggregation level, the

switches energy consumed by the access level, the

energy consumed by servers, and the total energy

consumed the whole system.

Figure 3: A summarized view of Cloud model parameters.

Table 1: The factors values per level.

Level

Energy

Management

Network

Switches

Virtualization Simulation Setup

Host DVFS

Hosts Dynamic

shutdown

Net Switch DVFS

Dynamic shutdown

Shutdown/ wake up

time

(

)

DVFS

VM static

configuration

Datacenter Load

Task Scheduler

A B C D E F G H I

1 Yes Yes Yes Yes 0.01 Yes Yes 0.3 Green

2 No No No No 0.05 No No 0.4 Round Robin

3 - - - - - - - 0.5 Best DENS

Level

Datacenter Topology Task Description

Topology

Core to

ation

Aggregation to

access

Access to host

Task size (MIPS)

Task memory

Task storage

Task description

size (bytes)

Task execution

deadline (s)

Task output

(Bytes)

J K L M N O P Q R S

1 T1 10 10 1 30 10 30 8.5 5 25

2 T2 100 15 5 35 15 35 9.0 10 30

3 T3 50 20 10 40 20 40 9.5 15 35

Where T1 is three-tier topology, T2 is three-tier high-speed and

T3 is three-tier heterogonous small

Moreover, these indicators make it possible to

evaluate the effectiveness of the various scheduling

algorithms. The first measure studied is the total

energy consumption by the physical resources of a

data center caused by the workload of the application.

the choice of model parameters including influencing

factors and key performance indicators was guided by

the possibilities allowed by the GreenCloud

simulator.

5 SIMULATION AND RESULTS

Through this article, we intend to evaluate the energy

efficiency of the Cloud model and to introduce a

solution of scheduling of the virtual machines while

respecting the energy efficiency aspect and the

response time aspect. In order to properly target the

actions to be undertaken, we proceed in the first place

to analyze the operation of the Cloud model through

the GreenCloud simulation platform. As announced,

we have carried out 36 simulations according to the

L36 Taguchi matric. During these experiments, we

have used three data center topologies. The first

topology used is a three-tier architecture which

includes 1536 servers organized into 512 racks as

Energy

Management

Network

Switches

Virtualization

Simulation

setup

Data center

Topo lo gy

Task

Description

Cloud

system

Simulation

duration

Switch Energy

(core)

Switch Energy

(agg.)

Switch Energy

(access)

Server Energy

Data center

load

Average Tasks

per server

Output

Input

Energy Efﬁciency Policies for Smart Digital Cloud Environment based on Heuristics Algorithms

359

well as 8 cores and 64 switches. The second topology

is a three-tier high speed which consists of 1536

servers and 2 cores and 256 switches. The last

topology applied is three-tier heterogeneous small

which consists of 288 servers organized in 48 racks,

2 cores, and 3 switches. Several conclusions have

emerged from the various simulations carried out (see

Table 3) and the analysis of the signal-to-noise ratio

(see Tables 4-7). Thereby, the first observation is that

the energy consumed depends strongly on the applied

topology.

Table 2: The Taguchi experience matrix L36.

Trials A B C D E F G H I J K L M N O P Q R S

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

2 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2

3 1 1 1 1 1 1 1 3 3 3 3 3 3 3 3 3 3 3 3

4 1 1 1 1 1 2 2 1 1 1 1 2 2 2 2 3 3 3 3

5 1 1 1 1 1 2 2 2 2 2 2 3 3 3 3 1 1 1 1

6 1 1 1 1 1 2 2 3 3 3 3 1 1 1 1 2 2 2 2

7 1 1 2 2 2 1 1 1 1 2 3 1 2 3 3 1 2 2 3

8 1 1 2 2 2 1 1 2 2 3 1 2 3 1 1 2 3 3 1

9 1 1 2 2 2 1 1 3 3 1 2 3 1 2 2 3 1 1 2

10 1 2 1 2 2 1 2 1 1 3 2 1 3 2 3 2 1 3 2

11 1 2 1 2 2 1 2 2 2 1 3 2 1 3 1 3 2 1 3

12 1 2 1 2 2 1 2 3 3 2 1 3 2 1 2 1 3 2 1

13 1 2 2 1 2 2 1 1 2 3 1 3 2 1 3 3 2 1 2

14 1 2 2 1 2 2 1 2 3 1 2 1 3 2 1 1 3 2 3

15 1 2 2 1 2 2 1 3 1 2 3 2 1 3 2 2 1 3 1

16 1 2 2 2 1 2 2 1 2 3 2 1 1 3 2 3 3 2 1

17 1 2 2 2 1 2 2 2 3 1 3 2 2 1 3 1 1 3 2

18 1 2 2 2 1 2 2 3 1 2 1 3 3 2 1 2 2 1 3

19 2 1 2 2 1 1 2 1 2 1 3 3 3 1 2 2 1 2 3

20 2 1 2 2 1 1 2 2 3 2 1 1 1 2 3 3 2 3 1

21 2 1 2 2 1 1 2 3 1 3 2 2 2 3 1 1 3 1 2

22 2 1 2 1 2 2 2 1 2 2 3 3 1 2 1 1 3 3 2

23 2 1 2 1 2 2 2 2 3 3 1 1 2 3 2 2 1 1 3

24 2 1 2 1 2 2 2 3 1 1 2 2 3 1 3 3 2 2 1

25 2 1 1 2 2 2 1 1 3 2 1 2 3 3 1 3 1 2 2

26 2 1 1 2 2 2 1 2 1 3 2 3 1 1 2 1 2 3 3

27 2 1 1 2 2 2 1 3 2 1 3 1 2 2 3 2 3 1 1

28 2 2 2 1 1 1 1 1 3 2 2 2 1 1 3 2 3 1 3

29 2 2 2 1 1 1 1 2 1 3 3 3 2 2 1 3 1 2 1

30 2 2 2 1 1 1 1 3 2 1 1 1 3 3 2 1 2 3 2

31 2 2 1 2 1 2 1 1 3 3 3 2 3 2 2 1 2 1 1

32 2 2 1 2 1 2 1 2 1 1 1 3 1 3 3 2 3 2 2

33 2 2 1 2 1 2 1 3 2 2 2 1 2 1 1 3 1 3 3

34 2 2 1 1 2 1 2 1 3 1 2 3 2 3 1 2 2 3 1

35 2 2 1 1 2 1 2 2 1 2 3 1 3 1 2 3 3 1 2

36 2 2 1 1 2 1 2 3 2 3 1 2 1 2 3 1 1 2 3

Indeed, the classification of the most influential

factors on energy efficiency shows that the topology

factor is ranked first for the four indicators studied. In

other words, it is essential to optimize the number of

switches and servers used to handle the different

requests of users. This first observation confirms the

proposals already announced in the first paragraphs

of this article. Moreover, the topology three-tier

heterogonous small is the most efficient on the

energetic level.

Table 3: The simulation results.

Trials

Key Performance Indicators (W*h)

Switch

Energy

(core)

Switch

Energy

(agg.)

Switch

Energy

(access)

Server

Energy

Total

Energy

1 466.10 932.10 1 375.70 3 541.00 6 314.90

2 112.30 493.50 1 480.70 4 001.10 6 087.60

3 59.20 118.50 10.50 425.50 613.70

4 537.20 1 074.40 1 585.80 3 725.40 6 922.80

5 1 033.40 458.50 1 375.70 3 542.30 6 409.90

6 55.30 110.60 9.80 375.50 551.20

7 1 112.30 493.50 1 480.70 3 510.90 6 597.40

8 59.20 118.50 10.50 372.40 560.60

9 466.10 932.10 1 375.70 3 792.40 6 566.30

10 59.20 118.50 10.50 349.60 537.80

11 466.10 932.10 1 375.70 3 542.30 6 316.20

12 1 112.30 493.50 1 480.70 3 698.90 6 785.40

13 51.40 102.80 9.10 309.20 472.50

14 501.60 1 003.30 1 480.70 3 598.50 6 584.10

15 1 191.20 528.50 1 585.80 4 212.50 7 518.00

16 55.30 110.60 9.80 329.30 505.00

17 537.20 1 074.40 1 585.80 3 869.70 7 067.10

18 1 033.40 458.50 1 375.50 3 784.50 6 651.90

19 501.60 1 003.30 1 480.70 3 510.00 6 495.60

20 1 191.20 528.50 1 585.80 3 861.60 7 167.10

21 51.40 102.80 9.10 355.50 518.80

22 1 112.30 493.50 1 480.70 3 510.00 6 596.50

23 51.40 102.80 9.10 341.60 504.90

24 501.60 1 003.30 1 480.70 3 998.50 6 984.10

25 1 112.30 493.50 1 480.70 3 378.00 6 464.50

26 59.20 118.50 10.50 372.60 560.80

27 466.10 932.10 1 375.70 3 789.50 6 563.40

28 1 033.40 458.50 1 375.70 3 159.70 6 027.30

29 55.30 110.60 9.80 352.40 528.10

30 537.20 1 074.40 1 585.80 4 213.60 7 411.00

31 51.40 102.80 9.10 312.80 476.10

32 501.30 1 003.30 1 480.70 3 754.70 6 740.00

33 1 191.20 528.50 1 585.80 4 213.00 7 518.50

34 537.20 1 074.40 1 585.80 3 691.40 6 888.80

35 1 033.40 458.50 1 375.70 3 540.90 6 408.50

36 55.30 110.60 9.80 375.50 551.20

CLOSER 2018 - 8th International Conference on Cloud Computing and Services Science

360

Furthermore, the energy consumed by the servers

keeps the most impact on the total energy consumed

by the data center, hence the obligation to reduce the

number of servers used for processing a batch of

queries. The variance analysis of the server energy

indicator (see Table 7 and Figure 4) indicates that the

topology is a highly influential factor based on the

value of the ratio P which is equal to zero.

Table 4: Factors ranks based on SNR Switch Energy (core).

Factor A B C D E F G H I J

Rank 17 13 15 14 19 18 16 7 3 1

Factor K L M N O P Q R S

Rank 8 10 12 5 11 6 9 2 4

Table 5: Factors ranks based on SNR Switch Energy (agg.).

Factor A B C D O F G H I J

Rank 17 18.5 16 13 4 15 14 8.5 6.5 1

Factor K L M N E P Q R S

Rank 5 11 3 6.5 18.5 10 12 2 8.5

Table 6: Factors ranks based on SNR Switch Energy (acc.).

Factor A B C D O F G H I J

Rank 15 18.5 13.5 17 7 13.5 16 3 9 1

Factor K L M N E P Q R S

Rank 8 11 10 6 18.5 4 12 2 5

Table 7: Factors ranks based on SNR Server Energy.

Factor A B C D E F G H I

Rank 13 5 7 11 6 14 10 2 4

Factor J k L M N O P Q R S

Rank 1 19 16 18 12 15 9 8 3 17

Figure 4: Main Effects Plot for SNR of Server Energy.

The regression analysis allowed us to define the

regression equation (10) of the energy server

indicator. This equation reflects the weight of each

factor in the performance and the impact of each input

on the value of the server energy indicator.

 = 5696.69−13.87+

50.26 +25.13 −6.49−52.38+

62.38−50.41+162.82−41.37−

1698.13−16.85−38.38+8.31+

14.0083 + 9.67 +23.63−50.74+

116.90−47.63

(10)

6 PROPOSED ALGORITHM

As depicted in Figure 5, we apply a three-tiered

architecture for the Cloud model studied. This

architecture relies on a combination of several

algorithms including ant colony optimization

(Ragmani et al., 2016b). Through the present article

and following analyses made in the previous

paragraph, we propose an algorithm for scheduling

virtual machines within Cloud by applying a

simulated annealing algorithm.

Figure 5: The proposed Cloud architecture.

Indeed, thanks to its interesting properties of

convergence, we target both to find an optimal

response in a reasonable time. In brief, the simulated

annealing (SA) is inspired by the Metropolis-

Hastings algorithm, which offers the possibility of

modeling a thermodynamic system. This algorithm

relies on a function to be minimized which refers to

the energy  in the real process and a temperature 

of the system. The variation of the temperature allows

to find out the different intermediary solutions before

reaching the optimal solution. This algorithm starts

with an initial state of the system that will be modified

to reach a new state. Two cases then arise; either the

new state improves the factor to be optimized or it

degrades it. The validation of the state which has

improved the factor to optimize allows us to find an

optimum in the vicinity of the initial state. On the

-64,55

-64,60

-64,65

-64,55

-64,60

-64,65

-64,55

-64,60

-64,65

-64,55

-64,60

-64,65

-64,55

-64,60

-64,65

-64,55

-64,60

-64,65

-64,55

-64,60

-64,65

321

-64,0

-64,5

-65,0

321

-64,5

-64,6

-64,7

321

-50

-60

-70

321

-64,58

-64,60

-64,62

321

-64,580

-64,605

-64,630

321

-64,58

-64,60

-64,62

321

-64,55

-64,60

-64,65

321

-64,56

-64,60

-64,64

321

-64,55

-64,60

-64,65

321

-64,55

-64,60

-64,65

321

-64,4

-64,8

-65,2

321

-64,58

-64,60

-64,62

Mean of SN ratios

B C D E

F G H I J

k L M N O

P Q R S

Energy Efﬁciency Policies for Smart Digital Cloud Environment based on Heuristics Algorithms

361

other hand, the acceptance of a bad state induces us

to seek an optimum outside the neighborhood of the

initial state (Chibante, 2010; Marotta et al., 2018).

There are two approaches to the variation of

temperature. The first approach is to keep the system

temperature constant. When the system reaches a

thermodynamic equilibrium, the temperature is

lowered. This approach corresponds to a resolution in

increments of temperature. The second approach is

based on a continuous decrease of the temperature

according to a precise law. The most applied one is

=, where  is less than 1. In both approaches,

the algorithm stops at a predefined temperature (see

Figure 6). In our case, we keep the second approach

because it allowed us to achieve better results. The

pseudo-code of the proposed scheduling algorithm is

summarized on Algorithm I. The proposed algorithm

is an application of simulated annealing. This

algorithm uses several inputs including the list of

virtual machines; the available servers, the number of

iterations and the initial temperature. The ultimate

goal is to plan the placement of virtual machines in

the data center servers to minimize power

consumption. The function to be minimized by the

algorithm is the total energy consumed. At each

iteration, the temperature is multiplied by a parameter

α which makes it possible to reduce the temperature.

In order to avoid a rapid convergence to a local

minimum, a random perturbation is applied which

consists in randomly modifying the location of the

machines in order to force the algorithm to look for

new possibilities. In order to validate the operation of

the proposed algorithm we have configured 22 virtual

machines and 5 servers and before testing several

combinations of parameters in order to identify the

best combination (see Table 8 and Figure 7).

Figure 6: The SA algorithme convergence for M=100,

N=50, T0=50, α= 0.87.

Table 8: SA parameters per combination.

Combination 1 2 3 4 5 6 7

M 100

N 50 50 50 50 50

T0 50 50 20 60

60 50

α 0.87 0.9 0.9 0.9 0.9 0.9 0.8

Figure 7: Best energy efficiency per configuration.

Algorithm I: SA Algorithm for VM Scheduling.

Inputs: VM list; hosts list;

N: Number iterations of first loop;

M: Number iterations of secondary loop;

: Initial Temperature; alpha: cooling

rate; G: Indicator of environmental

impact of data center (between 1 for good

data center and 10 for worst one);

Output: VM Allocation per server;

% Initialization

VM.affactation=random(Hosts)

% Replace by Best Solution identified

AllocatedHost= VM.affectation

T=T

;

% SA Iterations

For i=1to N

For j=1 to M

Set VM neighbor;

Apply random perturbation;

If AllocatedHostnew.Energy*G ≤

AllocatedHost.Energy

VM.affection = AllocatedHostnew;

else

delta= AllocatedHostnew.Energy

- AllocatedHost.Energy;

=(−/);

if p ≥ random

AllocatedHost = AllocatedHostnew;

End

% Save the best Energy Efficiency value

BestEnergyEfficency(i)=AllocatedHost

.Energy;

% Decrease temperature

T=alpha*T;

End

0 10 20 30 40 50 60 70 80 90 100

100

120

140

160

180

200

220

240

Iteration

Best Energy Efficiency

104

108

103

101

103

100

1234567

Best En ergy Number of Iteration

CLOSER 2018 - 8th International Conference on Cloud Computing and Services Science

362

7 CONCLUSIONS

In conclusion, through this article, we have pointed

out the importance of energy efficiency in the Cloud

computing environment. Then, we proceed to various

simulations based on the Taguchi concept in order to

evaluate deeply all major aspects of energy

consumption within a data center. Finally, we

introduced a scheduling proposal based on applying

an algorithm inspired by simulated annealing

algorithms in order to guarantee an efficient

scheduling strategy. We have proposed to integrate a

parameter relative to the source of the energy used in

a data center in order to give an advantage to data

center applying environmental standards. The main

characteristic of the proposed method is allowing to

model different responses according to influencing

factors. This knowledge is then used in the

optimization process of the studied system. In future

work, we propose to examine in more detail the

proposed solution within a real Cloud environment in

order to confirm its gain in terms of energy efficiency.

ACKNOWLEDGEMENTS

We express our deep thankfulness to the numerous

anonymous reviewers for their valuable comments.

REFERENCES

Aschberger, C., Halbrainer, F., 2013. Energy Efficiency in

Cloud Computing.

Beloglazov, A., Abawajy, J., Buyya, R., 2012. Energy-

aware resource allocation heuristics for efficient

management of data centers for Cloud computing.

Future Gener. Comput. Syst. 28, 755–768.

Canali, C., Lancellotti, R., Shojafar, M., 2017. A

Computation- and Network-Aware Energy

Optimization Model for Virtual Machines Allocation:

SCITEPRESS - Science and Technology Publications,

pp. 71–81.

Chibante, R., 2010. Simulated annealing, theory with

applications. Sciyo, Rijek, Crotia.

Chinnici, M., Quintiliani, A., 2013. An Example of

Methodology to Assess Energy Efficiency

Improvements in Datacenters. IEEE, pp. 459–463.

Guzek, M., Kliazovich, D., Bouvry, P., 2013. A holistic

model for resource representation in virtualized cloud

computing data centers, in: 2013 IEEE 5th

International Conference on Cloud Computing

Technology and Science (CloudCom), pp. 590–598.

Jin, H., Gao, W., Wu, S., Shi, X., Wu, X., Zhou, F., 2011.

Optimizing the live migration of virtual machine by

CPU scheduling. J. Netw. Comput. Appl. 34, 1088–

1096. https://doi.org/10.1016/j.jnca.2010.06.013.

Kliazovich, D., Arzo, S.T., Granelli, F., Bouvry, P., Khan,

S.U., 2013a. Accounting for load variation in energy-

efficient data centers, in: IEEE International

Conference on Communications (ICC), 2013. IEEE, pp.

2561–2566.

Kliazovich, D., Bouvry, P., Khan, S.U., 2013b. DENS: data

center energy-efficient network-aware scheduling.

Clust. Comput. 16, 65–75.

Kliazovich, D., Bouvry, P., Khan, S.U., 2012. GreenCloud:

a packet-level simulator of energy-aware cloud

computing data centers. J. Supercomput. 62, 1263–

1283. https://doi.org/10.1007/s11227-010-0504-1.

Li, D., Wang, W., Li, Q., Ma, D., 2013. Study of a Virtual

Machine Migration Method. IEEE, pp. 27–31.

https://doi.org/10.1109/CBD.2013.33.

Marotta, A., Avallone, S., 2015. A Simulated Annealing

Based Approach for Power Efficient Virtual Machines

Consolidation. Presented at the 2015 IEEE 8th

International Conference on Cloud Computing, IEEE,

New York, USA, pp. 445–452.

Marotta, A., Avallone, S., Kassler, A., 2018. A Joint Power

Efficient Server and Network Consolidation approach

for virtualized data centers. Comput. Netw. 130, 65–80.

Ragmani, A., El Omri, A., Abghour, N., Moussaid, K.,

Rida, M., 2016a. A global performance analysis

methodology: Case of cloud computing and logistics,

in: 3

International Conference (GOL), 2016. IEEE,

Fes Morocco, pp. 1–8.

Ragmani, A., El Omri, A., Abghour, N., Moussaid, K.,

Rida, M., 2016b. A performed load balancing algorithm

for public Cloud computing using ant colony

optimization, in: 2

International Conference

(CloudTech 2016). IEEE, Marrakech, Morocco, pp.

221–228.

Sallam, A., Li, K., 2014. A Multi-objective Virtual

Machine Migration Policy in Cloud Systems. Comput.

J. 57, 195–204. https://doi.org/10.1093/comjnl/bxt018.

Sarji, I., Ghali, C., Chehab, A., Kayssi, A., 2011.

CloudESE: Energy efficiency model for cloud

computing environments, in: 2011 International

Conference on Energy Aware Computing (ICEAC).

IEEE, pp. 1–6.

Sosinsky, B.A., 2011. Cloud computing bible. Wiley ; John

Wiley [distributor], Indianapolis, IN : Chichester.

Taguchi, G., Chowdhury, S., Wu, Y., Taguchi, S., Yano,

H., 2005. Taguchi’s quality engineering handbook.

John Wiley & Sons ; ASI Consulting Group, Hoboken,

N.J. : Livonia, Mich.

Velte, A.T., Velte, T.J., Elsenpeter, R.C., 2010. Cloud

computing a practical approach. McGraw-Hill, New

York.

Zhou, Z., Hu, Z., Song, T., Yu, J., 2015. A novel virtual

machine deployment algorithm with energy efficiency

in cloud computing. J. Cent. South Univ. 22, 974–983.

Energy Efﬁciency Policies for Smart Digital Cloud Environment based on Heuristics Algorithms

363