PEOPLE'S DEMOCRATIC REPUBLIC OF ALGERIA

MINISTRY OF HIGHER EDUCATION AND SCIENTIFIC RESEARCH

MOHAMED BOUDIAF UNIVERSITY - M’SILA

FACULTY : MATHEMATICS AND COMPUTER DOMAIN : MATHEMATICS AND COMPUTER

SCIENCE SCIENCE
DEPARTMENT : COMPUTER SCIENCE FILIERE : COMPUTER SCIENCE
N°:…………………………………………… OPTION : IDO

Thesis submitted for obtaining

Academic Master's degree
By: Hamza Sarra

Entitled

Job scheduling in cloud computing environment

Defended before the jury composed of:

……………………………... University of M’sila President

Dr. Barkat Abdelbasset University of M’sila Reporter
……………………………… University of M’sila Examiner

Academic year: 2019/2020

 Dedication

«To my husband, and my sons

Chihab Eddine, and Siradje Eddine »

ii
 Acknowledgment

Foremost, great thank for ALLAH, and I thank my Parents my sisters and brothers who

supported me during this academic journey and my whole life.

Then I would like to thank my thesis advisor Dr. Barkat Abdelbasset for the continuous support

of me, for his patience, motivation, enthusiasm, and immense knowledge. His guidance helped

me in all the time of research and writing of this thesis.

Many sincere thanks also go to mathematics and computer science Faculty members at

Mohamed Boudiaf University of M’sila for their insightful conversations and hard questions,

thank you for teaching me how to be a dedicated researcher.

Million thanks go to my responsible at work the vice rector of pedagogy Pr. Bouguerra Rabah

for his support and encouragement, and kindness during my study.

Special thanks to my uncle Younes for all his help to complete this work

Thank you very much.

HAMZA Sarra

August/2020

iii
 Abstract

Job scheduling in cloud computing environment

By: Hamza Sarra.

Supervisor: Dr. Barkat Abdelbaset.

Cloud computing provides a pool of virtualized computing resources and adopts pay-per-use

model. Task Scheduler is one of the most important cloud computing problems. And that cannot

be scheduled using single criteria, but according too many of the criteria's and rules. These rules

and criteria's to be agreed upon by the service provider and the user.

In this report, I present a particle swarm optimization (PSO) based scheduling algorithm to

minimize the completion execution time makespan. Experiment is conducted by varying

computation of tasks, number of particles and makespan in fitness function. Then I modify the

proposed PSO by adding a method for sorting the tasks by them length to execute them by

shortest job first (SJF).

The two proposed algorithm are implemented by using the CloudSim simulator. By comparing

the two proposed algorithms (PSO and SJF-PSO) the result given by (SJF-PSO) algorithm is

the lower Makespan.

Keywords:

Cloud Computing, scheduling algorithm, Particle Swarm Optimization (PSO), Shortest

Job First (SJF), Makespan, CloudSim.

iv
 TABLE OF CONTENTS

Introduction………………………………………………………………………………………... 1
I. Chapter I: Background and Literature Review
I.1. Overview…………………………………………………………………………………… 4
I.2. Cloud Computing…………………………………………………………………………... 4
I.2.1. Cloud Computing characteristics…………………………………………………….. 6
I.2.2. Cloud Computing Service Models…………………………………………………… 7
I.2.2.1. Software as a Service (SaaS)…………………………………………………… 8
I.2.2.2. Platform as a Service (PaaS)……………………………………………………. 8
I.2.2.3. Infrastructure as a Service (IaaS)…………………………………………...…... 8
I.2.3. Deployment Models………………………………………………………………….. 8
I.2.3.1. Private cloud………………………………………………………………......... 8
I.2.3.2. Community cloud…………………………………………………………......... 9
I.2.3.3. Public cloud……………………………………………………………….......... 9
I.2.3.4. Hybrid cloud………………………………………………………………......... 9
I.2.4. Cloud Computing Architecture………………………………………………………. 10
I.2.4.1. Cloud Provider…………………………………………………………….......... 10
I.2.4.2. Cloud Consumer………………………………………………………………... 10
I.2.4.3. Cloud Carrier…………………………………………………………………… 11
I.2.4.4. Cloud Broker…………………………………………………………………… 11
I.2.4.5. Cloud Auditor…………………………………………………………………... 11
I.3. Virtualization……………………………………………………………………………….. 11
I.3.1. Definition……………………………………………………………………………... 11
I.3.2. Approaches in Virtualization…………………………………………………………. 12
I.3.2.1. Full Virtualization…………………………………………………………......... 12
I.3.2.2. Paravirtualization……………………………………………………………….. 12
I.3.2.3. Hardware-Assisted Virtualization…………………………………………........ 13
I.3.3. Hypervisor and Its Role………………………………………………………………. 13
I.3.3.1. Types of hypervisor…………………………………………………………….. 13
I.3.4. Types of Virtualization……………………………………………………………….. 14
I.3.4.1. OS Virtualization……………………………………………………………….. 14

v
 I.3.4.2. Server Virtualization………………………………………………………......... 14
I.3.4.3. Memory Virtualization……………………………………………………......... 14
I.3.4.4. Storage Virtualization…………………………………………………………... 14
I.3.4.5. Network Virtualization…………………………………………………………. 15
I.3.4.6. Application Virtualization……………………………………………………… 15
I.3.5. Virtualization and Cloud Computing…………………………………………………. 15
I.4. Datacenter…………………………………………………………………………………... 17
I.4.1. What is a data center?.................................................................................................... 17
I.4.2. Information Technology Infrastructure of data center………………………………... 18
I.4.3. Cloud computing and data center……………………………………………………... 19
I.5. Conclusion…………………………………………………………………………………... 20
II. Chapter II: The Proposed PSO Scheduling Algorithm
II.1. Overview…………………………………………………………………………………… 21
II.2. Tasks scheduling in Cloud Computing……………………………………………………... 21
II.2.1. Task Scheduling Types in Cloud Computing………………………………………… 23
II.2.1.1. Job Scheduling…………………………………………………………………. 23
II.2.1.2. Workflow Scheduling………………………………………………………….. 23
II.2.1.3. VM Scheduling………………………………………………………………… 23
II.2.1.4. Storage Scheduling…………………………………………………………….. 24
II.2.2. Related works………………………………………………………………………… 25
II.3. PSO algorithm (particle swarm optimization)……………………………………………… 28
II.3.1. PSO Benefits…………………………………………………………………………. 30
II.4. Modeling job scheduling in Cloud Computing using PSO algorithm……………………… 31
II.5. Conclusion………………………………………………………………………………….. 36
III. Chapter III: Implementation and Experiment Result
III.1. Overview…………………………………………………………………………………... 37
III.2. Cloud computing simulation………………………………………………………………. 37
III.2.1. Reasons for adopting for modeling and simulation………………………………….. 38
III.3. Software Tools Used in the Research……………………………………………………… 38
III.3.1. Eclipse……………………………………………………………………………….. 38
III.3.2. CloudSim……………………………………………………………………………. 39
III.3.2.1. The pros of CloudSim Simulation…………………………………………….. 39
III.3.2.2. The cons of CloudSim simulation……………………………………………... 39

vi
 III.3.2.3. CloudSim Versions……………………………………………………………. 39
III.3.2.4. Architecture of CloudSim Simulation…………………………………………. 41
III.3.2.5. Basic scenario of CloudSim Simulation……………………………………….. 43
III.4. Implementation and Experiment Result…………………………………………………… 44
III.5. Conclusion…………………………………………………………………………………. 54
Conclusion…………………………………………………………………………………………… 55
Refereces……………………………………………………………………………………………... 56

vii
 LIST OF FIGURES

Fig I.1: Diagram showing overview of cloud computing……………………………….. 5

Fig I.2: The essential characteristics of cloud computing………………………………. 6

Fig I.3: Layers of the cloud service model and their respective responsibilities……….. 7

Fig I.4: Types of Cloud Deployment Models…………………………………………… 9

Fig I.5: NIST Cloud Computing Reference Architecture (CCRA)……………………... 10

Fig I.6: Virtualization overview………………………………………………………… 11

Fig I.6: Virtualized servers, storage, and network for cloud platform construction……. 17

Fig II.1: Resource allocation for task scheduling problem in cloud computing………… 22

Fig II.2: Task Scheduling Types in Cloud Computing………………………………….. 23

Fig II.3: Three main causes that may influence a bird’s (particle) trajectory…………… 29

Fig II.4: A sample particles……………………………………………………………… 33

Fig II.5: The flow chart of PSO for job scheduling in cloud computing………………... 35

Fig III.1: Cloud Computing Simulation frameworks……………………………………. 37

Fig III.2: Layered CloudSim architecture……………………………………………….. 41

Fig III.3: CloudSim class design diagram………………………………………………. 42

Fig III.3: Basic scenario of Simulation…………………………………………………. 43

viii
Fig III.4: CloudSim lifecycle…………………………………………………………… 44

Fig III.5: Class diagram of PSO scheduling……………………………………………. 45

Fig III.6: The InfBase Class……………………………………………………………... 46

Fig III.7 Experimental datacenter infrastructure………………………………………… 47

Fig III.8 Result of the Simulation shown the best makespane…………………………... 49

Fig III.9: creation of cloud entities……………………………………………………… 50

Fig III.10: execution statistics for 10 cloudlets with the proposed PSO………………… 50

Fig III.11: execution statistics for 20 cloudlets with the proposed SJF-PSO…………… 51

Fig III.12: Makespan comparison between PSO and SJF-PSO with respect of number
of Cloudlets……………………………………………………………………………… 53

Fig III.13: Makespan comparison between PSO and SJF-PSO with respect of number
of number of iterations………………………………………………………………….. 53

ix
 LIST OF TABLE

Tab II.1: Comparison between different tasks scheduling types in cloud computing…... 24

Tab II.2: The Pseudo code of PSO algorithm…………………………………………… 34

Tab III.1: Simulation Parameters………………………………………….…………….. 47

Tab III.2: pseudo code for the job scheduling simulation steps………………………… 48

Tab III.3: the result of the Makespan with respect to the number of iterations for PSO
and SJF-PSO…………………………………………………………………………….. 52

Tab III.4: the result of the Makespan with respect to the number of Cloudlets for PSO
and SJF-PSO…………………………………………………………………………….. 52

x
 I.
II.
III. Chapter III

IV. Introduction
INTRODUCTION

On Google, a search for the term “cloud computing” at join 2020 yields more than 400 million

hits, this is an indication of the importance of this technology in recent times, the Cloud

computing is derived from distributed computing, and it allows individuals and organizations

to access computing resources as services remotely through the Internet.

Cloud computing provides a pool of abstracted, virtualized resources, including computing

power, storage, platforms and software applications over the Internet based on users demand.

Due to its many characteristics such as elastic, scalable resource provision and cost-

effectiveness, cloud computing has been accepted by more and more users.

Cloud computing offers a great variety of services. Based on the level of services, there are

three categories generally. They are Infrastructure as a Service (IaaS), Platform as a Service

(PaaS), and Software as a Service (SaaS). IaaS puts servers, storage, networks, and data center

fabrics together as demanded by users. Cloud users can then install operating system and deploy

their own applications on the cloud. PaaS, on the other hand, provides middleware, database

and development tools. It enables users to deploy applications onto a virtualized cloud platform.

Finally, in SaaS the complete operating environment, along with applications, management,

and user interfaces, are provided to cloud users. Since all cloud services can be access by

subscription and run with a pay per-use model, cloud computing leverages many attractive

features to users, including low cost and simple management. There are many technical

challenges faced by cloud providers, such as maintaining high utilization while delivering

services that are low cost, short delay, and dynamic deployable. It is critical for cloud providers

to maintain an optimal tasks scheduling and management system to meet these challenges.

Just like any other software, cloud computing is not without issues. One of the most researched

issues today in cloud computing is task scheduling. Users submit their tasks to the cloud service
Introduction

provider at any given time and it is up to the cloud provider to execute. Cloud service providers

employ scheduling algorithms to ensure proper and efficient resource utilization on their end

and attempt to achieve customer satisfaction as well. The scheduler decides which resources

will be used, as well as which tasks will be executed on each of these resources, and allocate

tasks to the resources. The tasks scheduling problem, like general scheduling problems, is NP-

complete.

Tasks scheduling algorithms often utilize heuristics and metaheuristics, including soft

computing techniques, to obtain approximated solutions. In this report, I adopt a tasks

scheduling strategy using Particle Swarm Optimization (PSO). PSO, an applied soft computing

method developed by Kennedy and Eberhart, is one of the most advance evolutionary

algorithms driven from nature. PSO approximates an optimal solution by iteratively improving

a group of candidate solutions, called particles. Each particle is modified iteratively by the best

information from both the individual and the entire group. By collecting the cumulative

intelligence of whole group, the group is expected to move toward the most optimal solutions.

PSO works well on most global optimal problems. In addition, it is simple, effective, and of

low computational cost.

Makespan is a key performance measurement criterion assigned by cloud users and considered

by task schedulers. Makespan is the time from the beginning till the completion of the sequence

of tasks. Different application schedulers may use different policies with different objectives.

Some algorithms are designed to achieve minimum cost while others strive for minimum

makespan or for load balance. Most existing algorithms focus on achieving a single optimal

criterion. In this report, a job scheduling strategy to attain a minimal makespan is introduced.

The extensive research on all issues in real environment is extremely difficult because it

requires developers to consider network infrastructure and the environment, which may be

2
Introduction

beyond the control. Therefore, simulators are developed. To understand and apply better the

state-of-the-art of cloud computing simulators.

CloudSim is a simulator tool used to implement the task scheduling algorithms in a cloud

environment, using this framework, the cloud can be analyzed, designed, modeled, and

implemented for result analysis.

This thesis contains three main chapters; Chapter one presents the general idea of cloud

computing, its characteristics, services, deployment models and the architecture.

Will the Virtualization technology is the main foundation of Cloud Computing, for this reason

we will look in this chapter an overview of the virtualization, definition, Approaches,

Hypervisor, and types. Also, the section includes What defines Data Center and how cloud

computing uses it.

Chapter two explains the main concepts of scheduling cloud tasks, techniques on it. In addition

to, it presents samples of related work studies of various task scheduling algorithms in cloud

environment, namely: biogeographic based optimization (BBO), Tournament Selection genetic

algorithm (TS-GA), and the Ant colony optimization (ACO). Then we discuss the proposed

PSO methodology in details.

Chapter three we present a Simulation of proposed PSO algorithm for job scheduling in a cloud

environment using CloudSim. we declare the experimental settings and results.

Finally, we summarize the conclusion of this thesis and proposes some ideas for future work.

3
 I.
II.
III. Chapter I

IV. Background and Literature

Review
 Chapter I: Background and Literature Review

I. Chapitre1 : Clo ud Compu ting

I.1. Overview

Cloud Computing is one of the hottest terms in computing. Cloud computing is a leading
technique for bringing up a pool of computational resources with extendibility and employment
of these techniques by the usage of internet. It provides vast computing capability to big
companies which involve in dealing with enormous data produced daily on a pay and use
model.[15]

The key aim of this chapter is to conduct a comprehensive evaluation of the relevant literature
relating to the emerging field of Cloud Computing, focusing on three key areas of research
development in general: Cloud Computing technology and its characteristics; the virtualization
technology; and the data center.

The first section provides the definition of Cloud Computing, characteristics, service,
deployment models and the architecture.

The Virtualization technology is the main foundation of Cloud Computing, for this reason we
will look in the second section at an overview of the virtualization, definition, Approaches,
Hypervisor, and types.

The third section includes What defines Data Center and how cloud computing uses it.

I.2. Cloud Computing

Many practitioners in the commercial and academic spheres have attempted to define exactly
what “cloud computing” is and what unique characteristics it presents. Wikipedia have defined
it as follows: Cloud computing is the on-demand availability of computer system resources,
especially data storage and computing power, without direct active management by the user.
The term is generally used to describe data centers available to many users over the Internet.
Large clouds, predominant today, often have functions distributed over multiple locations from
central servers. If the connection to the user is relatively close, it may be designated an edge
server.

In Amazon Cloud computing is the on-demand delivery of IT resources over the Internet with
pay-as-you-go pricing. Instead of buying, owning, and maintaining physical data centers and
servers, you can access technology services, such as computing power, storage, and databases,
on an as-needed basis from a cloud provider like Amazon Web Services (AWS).

4
Chapter I: Background and Literature Review

The definition of cloud computing provided by NIST has gained significant traction within the
IT industry. According to this definition: Cloud Computing is a model for enabling convenient,
on-demand network access to a shared pool of configurable computing resources (e.g.,
networks, servers, storage, applications, and services) that can be rapidly provisioned and
released with minimal management effort or service provider interaction. This cloud model
promotes availability and is composed of five essential characteristics, three service models,
and four deployment models.[1]

A note to the definition says that “Cloud computing is still an evolving paradigm. Its definitions,
use cases, underlying technologies, issues, risks, and benefits will be refined in a spirited debate
by the public and private sectors. These definitions, attributes, and characteristics will evolve
and change over time.” This is undoubtedly true. Nevertheless, at this point in time, the NIST
definition is widely accepted and is increasingly regarded as authoritative. [5]

Fig I.1: Diagram showing overview of cloud computing, with typical types of applications
supported by that computing model.

5
Chapter I: Background and Literature Review

I.2.1. Cloud Computing characteristics

The five essential characteristics of the NIST definition are shown below (in Fig I.2) and
described in the following sections.

Fig I.2: The essential characteristics of cloud computing.

a. On-demand self-service: A consumer can unilaterally provision computing

capabilities, such as server time and network storage, as needed automatically without
requiring human interaction with each service’s provider. [4]
b. Broad network access: Capabilities are available over the network and accessed
through standard mechanisms that promote use by heterogeneous thin or thick client
platforms (e.g., mobile phones, laptops, and personal digital assistants). [4]
c. Elastic resource pooling: The provider’s computing resources are pooled to serve
multiple consumers using a multitenant model, with different physical and virtual
resources dynamically assigned and reassigned according to consumer demand. There
is a sense of location independence in that the customer generally has no control or
knowledge over the exact location of the provided resources but may be able to specify
the location at a higher level of abstraction (e.g., country, state, or data center).
Examples of resources include storage, processing, memory, and network bandwidth.
[4]

6
Chapter I: Background and Literature Review

d. Rapid elasticity: Capabilities can be rapidly and elastically provisioned, in some cases
automatically, to quickly scale out and rapidly released to quickly scale in. To the
consumer, the capabilities available for provisioning often appear to be unlimited and
can be purchased in any quantity at any time. [4]
e. Measured service: Cloud systems automatically control and optimize resource use by
leveraging a metering capability at some level of abstraction appropriate to the type of
service (e.g., storage, processing, bandwidth, and active user accounts). Resource usage
can be monitored, controlled, and reported providing transparency for both the provider
and consumer of the utilized service.[4]
I.2.2. Cloud Computing Service Models

There are the following three main deferent types of service models for cloud computing
defined by [1]:

Fig I.3: Layers of the cloud service model and their respective responsibilities.[6]

7
Chapter I: Background and Literature Review

I.2.2.1. Software as a Service (SaaS)

The capability provided to the consumer is to use the provider’s applications running on a cloud
infrastructure. The applications are accessible from various client devices through either a thin
client interface, such as a web browser (e.g., web-based email), or a program interface. The
consumer does not manage or control the underlying cloud infrastructure including network,
servers, operating systems, storage, or even individual application capabilities, with the possible
exception of limited user-specific application configuration settings. [1]

I.2.2.2. Platform as a Service (PaaS)

The capability provided to the consumer is to deploy onto the cloud infrastructure consumer-
created or acquired applications created using programming languages, libraries, services, and
tools supported by the provider. The consumer does not manage or control the underlying cloud
infrastructure including network, servers, operating systems, or storage, but has control over the
deployed applications and possibly configuration settings for the application-hosting environment. [1]

I.2.2.3. Infrastructure as a Service (IaaS)

The capability provided to the consumer is to provision processing, storage, networks, and other
fundamental computing resources where the consumer is able to deploy and run arbitrary
software, which can include operating systems and applications. The consumer does not
manage or control the underlying cloud infrastructure but has control over operating systems,
storage, and deployed applications; and possibly limited control of select networking
components (e.g., host firewalls). [1]

I.2.3. Deployment Models

Apart from the service models, NIST [1] defined the following deploying models for the cloud:

I.2.3.1. Private cloud

The cloud infrastructure is provisioned for exclusive use by a single organization comprising
multiple consumers (e.g., business units). It may be owned, managed, and operated by the
organization, a third party, or some combination of them, and it may exist on or off premises.
[1]

8
Chapter I: Background and Literature Review

I.2.3.2. Community cloud

The cloud infrastructure is provisioned for exclusive use by a specific community of consumers
from organizations that have shared concerns (e.g., mission, security requirements, policy, and
compliance considerations). It may be owned, managed, and operated by one or more of the
organizations in the community, a third party, or some combination of them, and it may exist
on or off premises.[1]

I.2.3.3. Public cloud

The cloud infrastructure is provisioned for open use by the general public. It may be owned,
managed, and operated by a business, academic, or government organization, or some
combination of them. It exists on the premises of the cloud provider.[1]

I.2.3.4. Hybrid cloud

The cloud infrastructure is a composition of two or more distinct cloud infrastructures (private,
community, or public) that remain unique entities, but are bound together by standardized or
proprietary technology that enables data and application portability (e.g., cloud bursting for
load balancing between clouds).[1]

Fig I.4: Types of Cloud Deployment Models [7]

9
Chapter I: Background and Literature Review

I.2.4. Cloud Computing Architecture

The cloud architecture consists of five separate components which work together to provide on-
demand services.

Figure I.5 is taken from National Institute of Standards and Technology (NIST) Cloud
computing reference architecture [2]. defines five major cloud actors: cloud provider, cloud
consumer, cloud carrier, cloud auditor and cloud broker. These are briefly described as further.

Fig I.5: NIST Cloud Computing Reference Architecture (CCRA) [2]

I.2.4.1. Cloud Provider

The cloud provider are organizations that provides cloud services. The cloud provider has
control over the IT infrastructure and manages the technical outages if planned or unplanned.
Cloud provider also ensures that agreed Service Level Agreements (SLA’s) are achieved. [3]

I.2.4.2. Cloud Consumer

A cloud consumer is a person(s) or an organization using cloud service(s) and having an

agreement with cloud provider or cloud broker. [3]

10
Chapter I: Background and Literature Review

I.2.4.3. Cloud Carrier

The cloud carriers are networks and telecommunication companies, which ensure that services
from cloud provider are available to cloud consumers. Cloud carrier work closely with cloud
provider to meet agreed SLA’s. [3]

I.2.4.4. Cloud Broker

The cloud brokers are third party companies, which work closely with both cloud providers and
cloud consumers. Generally, these are consulting companies and so they can easily sell various
cloud solutions to their existing customers as well as to new customers. [3]

I.2.4.5. Cloud Auditor

The cloud auditors are third parties who are specialized in independent assessment of cloud
services offered by cloud providers. A cloud auditor can audit various areas such as security,
privacy, performance, licensing, operations and other areas to highlight the gaps against various
operations and data privacy standards. [3]

I.3. Virtualization

I.3.1. Definition

Virtualization could be defined as separation of logical operations from their physical

environment. To give an easy example on virtualization, we can use virtual memory. Virtual
memory is an extension to system memory in a computer, which is used from the hard drive,
rather than from the RAM memory.[8]

If the system memory runs out, this virtual extension can be used to keep the system running.

Fig I.6: Virtualization overview.[8]

11
Chapter I: Background and Literature Review

I.3.2. Approaches in Virtualization

There have been many approaches adopted in the implementation of virtualization technology.
Some of the important approaches are discussed in the following subsections.[4]

I.3.2.1. Full Virtualization

Full virtualization uses a special kind of software called a hypervisor. The hypervisor interacts
directly with the physical server’s hardware resources, such as the CPU and storage space, and
acts as a platform for the virtual server’s OSs. It helps to keep each virtual server completely
independent and unaware of the other virtual servers running on the physical machine.

Each guest server or the virtual machine (VM) is able to run its own OS. That means one virtual
server could be running on Linux and the other one could be running on Windows. Examples
include VMWare ESX and VirtualBox. In the full virtualization, the guest OS is unaware of the
underlying hardware infrastructure. That means the guest OS is not aware of the fact that it is
running on a virtualized platform and of the feeling that it is running on the real hardware. In
this case, the guest OS cannot communicate directly to the underlying physical infrastructure.

The OS needs the help of virtualization software hypervisors to communicate with the
underlying infrastructure. The advantages of the full virtualization include isolation among the
various VMs, isolation between the VMs and the hypervisor, concurrent execution of multiple
OSs, and no change required in the guest OS. A disadvantage is that the overall system
performance may be affected due to binary translation. [4]

I.3.2.2. Paravirtualization

In this case, VMs do not simulate the underlying hardware, and this uses a special API that a
modified guest OS must use. Examples include Xen and VMWare ESX server. In this type of
virtualization, partial simulation of the underlying hardware infrastructure is achieved. This is
also known as partial virtualization or OS-assisted virtualization. This virtualization is different
from the full virtualization in that, here, the guest OS is aware of the fact that it is running in a
virtualized environment. In this case, hypercalls are used for the direct communication between
the guest OS and the hypervisor.

In paravirtualization, a modified or paravirtualized guest OS is required. 107 Technological

Drivers for Cloud Computing An advantage of this approach is that it improves the overall
system performance by eliminating the overhead of binary translation. A disadvantage could
be that a modification of the guest OS is required. [4]
12
Chapter I: Background and Literature Review

I.3.2.3. Hardware-Assisted Virtualization

In this type of virtualization, hardware products supporting the virtualization are used.
Hardware vendors like Intel and AMD have developed processors supporting the virtualization
through the hardware extension. Intel has released its processor with its virtualization
technology VT-x, and AMD have released its processor with its virtualization technology
AMD-v to support the virtualization. An advantage of this approach could be that it eliminates
the overhead of binary translation and paravirtualization. A disadvantage includes the lack of
support from all vendors. [4]

I.3.3. Hypervisor and Its Role

The concept of using VMs increases the resource utilization in a cloud computing environment.
Hypervisors are software tools used to create the VMs, and they produce the virtualization of
various hardware resources such as CPU, storage, and networking devices. They are also called
virtual machine monitor (VMM) or virtualization managers. They help in the virtualization of
cloud data centers (DCs). The various hypervisors used are VMware, Xen, Hyper-V, KVM,
etc.

Hypervisors help to run multiple OSs concurrently on a physical system sharing its hardware.
Thus, a hypervisor allows multiple OSs to share a single hardware host. In this case, every OS
appears to have the host’s processor, memory, and other resources allocated solely to it.
However, the hypervisor is actually controlling the host processor and resources and in turn
allocates what is needed to each OS.

The hypervisor also makes sure that the guest OSs (called VMs) do not interrupt each other. In
virtualization technology, hypervisor manages multiple OSs or multiple instances of the same
OS on a single physical computer system. Hypervisors are designed to suit a specific processor,
and they are also called virtualization managers. [4]

I.3.3.1. Types of hypervisor

Type 1 hypervisor: This type of hypervisor runs directly on the host computer’s hardware in
order to control the hardware resources and also to manage the guest OSs. This is also known
as native or bare-metal hypervisors. Examples include VMware ESXi, Citrix XenServer, and
Microsoft Hyper-V hypervisor. [4]

13
Chapter I: Background and Literature Review

Type 2 hypervisor: This type of hypervisor runs within a formal OS environment. In this type,
the hypervisor runs as a distinct second layer while the guest OS runs as a third layer above the
hardware. This is also known as the hosted hypervisors. Examples include VMware
Workstation and VirtualBox.108 Essentials of Cloud Computing. [4]

I.3.4. Types of Virtualization

Depending on the resources virtualized, the process of virtualization can be classified into the
following types.

I.3.4.1. OS Virtualization

In OS virtualization, a desktop’s main OS is moved into a virtual environment. The computer

that is used by the service consumers remains on their desk, but the OS is hosted on a server
elsewhere. Usually, there is one version of the OS on the server, and copies of that individual
OS are given to the individual user. Various users can then modify the OS as they wish, without
affecting the other users. [4]

I.3.4.2. Server Virtualization

In server virtualization, existing physical servers are moved into a virtual environment, which
is then hosted on a physical server. Modern servers can host more than one server
simultaneously, which allows the users to reduce the number of servers to be reserved for
various purposes. Hence, IT and administrative expenditures are reduced. Server virtualization
can use the virtual processors created from the real hardware processor present in the host
system. The physical processor can be abstracted into a collection of virtual processors that
could be shared by the VMs created. [4]

I.3.4.3. Memory Virtualization

In main memory virtualization, the virtual main memory that is abstracted from the physical
memory is allocated to various VMs to meet their memory requirements. The mapping of
physical to virtual memory is performed by the hypervisor software. The main memory
virtualization support is provided with the modern x86 processors. Also, the main memory
consolidation in the virtualized cloud DCs could be performed by the hypervisor by aggregating
the free memory segments of various servers to create a virtual memory pool that could be
utilized by the VMs. [4]

14
Chapter I: Background and Literature Review

I.3.4.4. Storage Virtualization

In storage virtualization, multiple physical hard drives are combined into a single virtualized
storage environment. To various users, this is simply called cloud storage, and it could be a
private storage, such that it is hosted by a company, or a public storage, such that it is hosted
outside of a company like DropBox, or a mixed approach of the two.

In the case of storage virtualization, physical storage disks are abstracted to a virtual storage
media. In cloud DCs, high availability and backup of the user’s data are achieved 109
Technological Drivers for Cloud Computing through storage virtualization technology. Modern
hypervisors help in achieving storage virtualization. The concept of storage virtualization is
implemented in advance storage techniques such as storage area networks (SANs) and network-
attached storage (NAS). [4]

I.3.4.5. Network Virtualization

In network virtualization (NV), logical virtual networks are created from the underlying
physical network. The physical networking components such as the router, switch, or network
interface card could be virtualized by the hypervisor to create logical equivalent components.
Multiple virtual networks can be created by using the same physical network components that
could be used for various purposes. NV can also be achieved by combining various network
components from multiple networks. [4]

I.3.4.6. Application Virtualization

In application virtualization, the single application installed on the central server is virtualized
and the various virtualized components of the application will be given to the users requesting
the services. In this case, the application is given its own copy of components such as own
registry files and global objects that are not shared with others. The virtual environment pre-
vents conflicts in the resource usage. An example is the Java Virtual Machine (JVM).

In the cloud computing environment, the CSPs deliver the SaaS model through the application
virtualization technology. In the case of application virtualization, the cloud users are not
required to install the required applications on their individual systems. They can, in turn, get
the virtualized copy of the application, and customize and use it for their own purposes. [4]

15
Chapter I: Background and Literature Review

I.3.5. Virtualization and Cloud Computing

The Virtualization technology is the main foundation of Cloud Computing offerings [9].

This concept refers to the set of hardware and software tools enabling the abstraction of a
physical resource into several logical units, that can be used separately by different Operating
Systems (OS) and applications [10]. Resources Virtualization is a key feature for Cloud
providers to build efficient, flexible and cost-effective computing models satisfying the Cloud
market challenges. It enables simple resource management and dynamic resource resizing,
reduced hardware costs due to resource sharing, isolation and fairness between tenants,
increased availability and quick recovery through easy backups and rapid migrations [11, 12].

Virtualization of servers on a shared cluster can consolidate the web services. As the virtual
machines are the container of cloud services, the provisioning tools will first find the
corresponding physical machines and deploy the virtual machines to those nodes before
scheduling the service to run on the virtual nodes. In cloud computing, virtualization is not only
means the system virtualization platforms are uses but also some fundamental services are
provided.[16]

In addition, in cloud computing, virtualization also means the resources and fundamental

infrastructure is virtualized. From the user point of view, the user will not care about the
computing resources that are used for providing the services. Cloud users do not need to know
and have no way to discover physical resources that are involved while processing a service
request. And also, from the developer point of view, the application developers are not care
about some infrastructure issues such as scalability, fault tolerant i.e. they are virtualized.
Application developers focus on the service logic.[16]

System Virtualization: In many cloud computing systems, system virtualization software is

used. System virtualization software is a special kind of software which simulates the execution
of hardware and run even the unmodified operating systems. Cloud computing systems use the
virtualization software as the running environment for legacy software such as old operating
systems and unusual applications. Virtualization software is also used as the platform for
developing new cloud applications that the developers can use any operating systems and
programming environments as they like. The development environment and deployment
environment can now be the same which eliminates some runtime troubles. The system
virtualization supported is illustrated in Fig I.6 with a virtualization infrastructure.[16]

16
Chapter I: Background and Literature Review

Fig I.6: Virtualized servers, storage, and network for cloud platform construction.[16]

I.4. Datacenter
I.4.1. What is a data center?

Data center can be defined as (Rihards Balodis and Inara Opmane ): A data center is a physical
environment facility intended for housing computer systems and associated components. Data
centers comprise the above-mentioned computer systems and staff that maintains them. The
necessary physical environment facility encompasses power supplies with the possibility to
ensure backup power. necessary communication equipment and redundant communication
cabling systems, air conditioning, fire suppression and physical security devices for staff
entrances.

Another way data center can be defined as Data center infrastructure comprises a large number
of servers interconnected by a network.[13]

17
Chapter I: Background and Literature Review

I.4.2. Information Technology Infrastructure of data center

Data center IT equipment may be classified as servers, storage, and networking devices. Servers
are mounted within racks and consist of hardware resources (such as CPUs, NICs, and RAMs)
hosting applications like video streaming and big data.

All the data generated by these applications are stored in storage systems. Data center storage
consists of high capacity (around Terabytes) disk drives or flash devices. The storage tiers are
connected to the servers either directly or through networking devices (managed by a global
distributed file system), forming a NAS (Network Attached Storage). The RAID (Redundant
Array of Independent Disks) storage technology can be used to provide high availability,
redundancy, and increase fault tolerance.

Networking equipment manages the internal communication between servers and storage
systems as well as all the input/output data flow from/to the data center. Typically, a data center
network is designed based on three hierarchical levels: core, distribution, and edge. It is through
the core level that all data center traffic (ingress/egress) to the outside world will be managed.
Distribution is a level between the edge and core levels which aggregates the edge switches. Its
main goal is simply network management and cabling reduction. Finally, the edge level passes
data from devices (generating or consuming data) to the distribution or core levels.

Manage all these components is a great challenge because hardware clusters have a deep and
complex memory/storage hierarchy, the system is composed of heterogeneous components, and
there are many failure-prone components. It is necessary to use an additional software layer to
provide an abstraction of this variety of components. According to, this software layer has three
levels: platform level, cluster level, and application level. The platform-level software is
composed of firmware, operational system, and kernel that provide an abstraction of hardware
of a single machine in the cluster and provide server-level services. The cluster-level software
is related to any software that managed resources at cluster level, such as distributed file
systems, management resource systems, and monitoring systems. The application level
software is composed of software that implements a particular service. The set of all software
used to monitor, measure, manage and control the data center behavior is named DCIM (Data
Center Infrastructure Management).[17]

18
Chapter I: Background and Literature Review

I.4.3. Cloud computing and data center

The term “cloud” started emerging around the same time wireless handheld devices started to
become popular in the marketplace. When accessing the web via a wireless handheld device, it
seems like you are pulling data out of the clouds. It is natural, then, that the data centers
providing this information should be called cloud data centers. Today, it appears that everyone
is jumping on the “cloud” bandwagon with all kinds of cloud companies, cloud products, and
cloud services entering the market.

Cloud data centers are being rapidly deployed around the world. Since these installations must
support up to hundreds of thousands of servers, data center efficiency and cost of operations
have become critical. Because of this, some cloud data centers have been erected near cheap
electrical power sources, such as hydroelectric dams, or in colder climates to help reduce
cooling costs. Some companies, such as Microsoft®, are building modular data centers using
pods, which are self-contained server storage and networking modules the size of a shipping
container. These modules are trucked in, stacked up, and connected to power, cooling, and
networking. Other data centers use server racks as the basic building block and contain rows
and rows of these racks. No matter what the structure, networking is an important part of these
large cloud data center networks.[14]

19
Chapter I: Background and Literature Review

I.5. Conclusion

In this chapter the fundamental aspects and the concepts which are present, while facing the
fields of cloud computing and virtualization are the object of giving an illustration of the recent
techniques and the new concepts towards which the data processing is directed. But always the
optimization will remain the first objective, searching to optimize the exploitation of the
available resources in a cloud or more precisely searching to tasks scheduling in a cloud is an
important research topic in cloud computing, a lot of work which consists to implement
optimization algorithms are realized. In the next chapter, we first present some work and then
we try to demonstrate and implement a proposed PSO algorithm using the CloudSim platform.

20
 I.
II.
III. Chapter II

IV. The Proposed PSO

Scheduling Algorithm
Chapter II: The Proposed PSO Scheduling Algorithm

II.1. Overview

Cloud Computing provides on demand access to some shared resources and services. These
services are provided as a service over a network, and can be accessed over the internet.

Cloud computing has become an interesting and beneficial way of changing the whole
computing. in the cloud computing the users submit their tasks so as to be executed as soon as
possible in the available shared pool of resource. In each cloud provider datacenter, there is a
component that control and manage the tasks scheduling in different VMs. Schedulers for cloud
computing determine on which processing resource jobs of a workflow should be allotted.

Task scheduling in cloud computing is an NP-complete problem [18]. There are no specific
methods to get polynomial time solutions for these problems. The meta-heuristics-based
technique have gained attention to get near optimal solutions for these complex problems.

This chapter is devised in three main section, in the first one we describe the tasks scheduling
in cloud computing, and his different types , then the second section we presents a review study
of various task scheduling algorithms in cloud environment, based on meta-heuristics
techniques namely: biogeographic based optimization (BBO), Tournament Selection genetic
algorithm (TS-GA), and the Ant colony optimization (ACO), in the last section a case study on
Practices swarm optimization (PSO) is designed. The PSO algorithm has been tested using
CloudSim toolkit.

II.2. Tasks scheduling in Cloud Computing

Scheduling is a process to determine the order resource which is mapped to be executed. The
main advantage of scheduling algorithm represented a high performance. It is one of the
important activities performed in grid and cloud computing environments, as its objective is to
deliver available computing resources as services over networks [18]. On the other hand, the
task scheduling algorithm is considered a complex process because it must schedule a large
number of tasks into the available resources. Also, there are many parameters that should be
taken into consideration to develop a task scheduling algorithm. Some of these parameters are
important from the Cloud user perspective (i.e., tasks compilation time, cost, and response
time). Other parameters are important from the Cloud provider perspective (i.e., resource
utilization, fault tolerant, and power consumption). [19]

21
Chapter II: The Proposed PSO Scheduling Algorithm

Tasks scheduling It consists of assigning a set of Tasks T to all available Resources R in

Datacenters, in order to complete all tasks under constraints imposed, effectively satisfy user U
requests and maximize the revenue of the provider [20], as shown in Fig II.1.

Fig II.1: Resource allocation for task scheduling problem in cloud computing [20]

The broker can be a manageable server or a specific datacenter in which the scheduling policy
locates. Whereas, the resources are available at the datacenter level. The Cloud provides
services to the user using three main steps as illustrated in the Fig II.1:

• The user sends requests.

• The broker receives and analyzes these requests.

• Application of the scheduling approaches to affect each task to the appropriate

resource.[20]

22
Chapter II: The Proposed PSO Scheduling Algorithm

II.2.1. Task Scheduling Types in Cloud Computing

The diversity of services providing (application, storage, CPU, etc.) and the necessity to make
it available to users, as well as the need for a better management of datacenters, led to a
multiplicity of task scheduling [20] as we illustrate in Fig II.2.

Fig II.2: Task Scheduling Types in Cloud Computing. [20]

II.2.1.1. Job Scheduling

Usually, job calls on a set of independent tasks [3]. Most approaches begin by evaluating the
weight of the tasks in its pool according to (waiting time, deadline, arrival time, execution time,
etc.) to determine the priority of each task in order to estimate the best mapping.[20]

II.2.1.2. Workflow Scheduling

Set of dependent tasks, represented with Directed Acyclic Graph (DAG) where each node
indicates the tasks and each edge represents the communication between dependency tasks. The
task can only be executed once all its parent tasks have been completed. [20]

II.2.1.3. VM Scheduling

Clouds providers create groups of VMs in a resource pool, then each VM should assign to a
host. It is one of the main challenges in implementing IaaS because there are many constraints
defined in a Service Level Agreement (SLA) document which explain how to associate the
VMs with each client. This latter problem can be solved by a VM migration technique. [20]

23
 Chapter II: The Proposed PSO Scheduling Algorithm

II.2.1.4. Storage Scheduling

We mean a set of data block, often with large capacity and heterogeneous type (image, text,
sound, etc.) which are stored in clusters that exist in different geographical places. Also, its size
increases continually as is the case in the bigdata. Furthermore, to obtain a good optimization,
we require a high processing method for storing and retrieving data, so scheduling is considered
as helpful solution to deal with this type of problems. For this purpose, it is generally necessary
to create a partition of data, based on the available information about storage sites and
resources.[20]

Task type Main scheduling phases Main parameters Example of queries

Job • Resource allocation • Weight tasks or size • Job request

phase(allocated pool of tasks
• Web applications
to set of resources)

Workflow • Tasks sorting phase • Data transmission • Workflow

application
• Resource allocation • Bandwidth
phase(allocated pool of tasks
• Sub-deadline of
to set of resources)
tasks

VM • VMs placement (allocated set • Consolidation of • VM requests

of VMs to set pf physical VMs
resources)
• Migration of VMs

Storage • Distribution (location) of data • Chunka • Big data application

(allocated blocks of data to
• Storage requests
cluster resources)
• Data mining
application

Tab II.1: Comparison between different tasks scheduling types in cloud computing [20]

24
Chapter II: The Proposed PSO Scheduling Algorithm

II.2.2. Related works

The job scheduling problem for the optimization objectives of minimizing the maximum
completion time has been proved to be NP-complete problems. So for this reason it is basically
handled by various known heuristic methods which give solutions for problem instances
basically in restricted manner. [21]

Bio inspired algorithm help us to solve the problems of optimization. Many researchers did
enormous work in this area from the past few decades. However, still there is a large more
scope for bio inspired algorithm in exploring new application and opportunities in cloud
computing, and they were used to tackle various challenges faced in Cloud Computing tasks
scheduling. [22]

The first interest work is [18] where the authors have used the biogeographic based optimization
(BBO), they formalize the problem as Follow:

A cloud node (CN) is a set of computational resources with limited computation speed on the
cloud, CN = (C1,C2,….Cm), each CN has corresponding calculating speed.

The set of CN speed is S =(s1, s2, . . . , sm). The unit of si =(1, 2, . . . ,m) is the number of Cycles
Per Unit Time si is the speed of CN Ci. A set of jobs J = (J1,J2,….Jn), each job has a length L =
(l1,l2,….ln). The authors introduce the matrix A=(aij) (i = 1, 2, . . . , m; j = 1, 2, . . . , n) as the
𝐿𝑗
time it takes CN Ci to complete job Jj. the time taking by a node i to execute a job j, 𝑎𝑖𝑗 = 𝑆𝑖

They define the decision matrix X = (xij) (i = 1, 2, . . . ; m; j = 1; 2; . . . n) tel que :

𝟏 𝒊𝒇 𝑱𝒋 𝒊𝒔 𝒂𝒔𝒔𝒊𝒈𝒏𝒆𝒅 𝒕𝒐 𝒄𝒍𝒐𝒖𝒅 𝒏𝒐𝒅𝒆 𝑪𝒊

𝒙𝒊𝒋 = {
𝟎 𝒐𝒕𝒉𝒆𝒓𝒘𝒊𝒔𝒆

For a job assignment on the cloud nodes, two essential parameters are calculated to measure

the performance of this assignment; The first is the flowtime for each cloud node which is equal

to the time taken by this cloud node to execute all of the tasks assigned to it, for example for

the cloud node Ci the flowtime

𝑺𝑭(𝑪𝒊 ) = ∑ 𝒂𝒊𝒋 ∗ 𝒙𝒊𝒋

𝒋=𝟏

25
Chapter II: The Proposed PSO Scheduling Algorithm

The second parameter is the makespan (MS), which is equal to the maximum flowtime of cloud

nodes

𝑴𝑺 = 𝒎𝒂𝒙𝒊∈𝟏..𝒎 𝑺𝑭(𝑪𝒊 )

Based on this modeling, the authors propose an optimization based on the biogeography

algorithm for optimal planning of tasks in the cloud, where the objective here is to minimize

the makespan of the X matrix.

The main goal of scheduling in the second article [19] is to enhance the completion time for
executing all tasks on the VMs, in the same time, minimize the total cost of usage the resource
and maximize utilization of the resource. For this reason, the authors propose the Tournament
Selection-Genetic Algorithm (TS-GA) which is task scheduling algorithm based on the default
GA with some modifications. According to these modifications the parents will be considered
in each population beside the produced child after the crossover process. Also, the Tournament
Selection is used to select the best chromosomes to overcome the limitation of the population
size. So the main idea of this proposed algorithm is that after each selection in the population,
there is a solution that might satisfy good fitness function, but it is not selected to crossover
process.

By the proposed algorithm, this solution is not removed from the population, but it is chosen
and added to the population when next iteration is started. This step is considered as a good step
as some of the iterations can generate the best solution.

The representation of solutions in task scheduling for each chromosome consists of VM ID and
ID for each task to be executed on these VM. Each VM and the executed tasks on it are encoded
into the binary bit.

The completion time of task Ti on VMj as is defined using equation Completion

time=CTmax[i,j]

where

Therefore, to reduce the completion time which can be denoted as CTmax, the execution time
of each task for each virtual machine must be calculated for the scheduling purpose. If the

26
Chapter II: The Proposed PSO Scheduling Algorithm

processing speed of virtual machine VMj is PSj, then the processing time for task Pi can be
calculated by equation

Where, Pij the processing time for task Pi on VMj and Ci computational complexity of task Pi

The processing time of each task in the virtual machine can be calculated by equation

Tournament Selection is used to overcome the limitation of the population size. Two individuals
are chosen at random from the population. A random number r is then chosen between 0 and 1.
If r < k (where k is a parameter, for example, 0.75), the fitter of the two individuals is selected
to be a parent; otherwise the less fit individual is selected. The non-chosen individuals are then
returned to the original population and could be selected again.

Then a new crossover has been used differently from the used crossover in the original GA.
Therefore, two chromosomes which are selected to crossover process to generate two offspring
will be considered as offspring also. So four children produced. After that, the two best children
are chosen from these.

In this modified TS-GA algorithm the new populations after crossover are added into old
populations (parents).

There is a solution that might satisfy good fitness function, but it is not selected during the
crossover process. By the proposed TS-GA algorithm, this solution is not removed from the
population, but it is chosen and added to the population when next iteration is started. This step
is considered as good step as some of the iterations can generate the best solution.

The authors of the 3rd article[27] proposed a Multi-objective Optimization Algorithm for Cloud
Computing Task Scheduling Based on Improved Ant Colony Algorithm (MO-ACO) to find the
optimal resource allocation for each task in the dynamic cloud System which minimizes the
makespan and costs of tasks on the entire system, and balance the entire system load.

In the algorithm, each ant looks for the appropriate virtual machine for the corresponding task.
The ants finally find an optimal matching scheme according to the optimized target. The ants
find the optimal solution in parallel, and communicate with each other through information.

27
Chapter II: The Proposed PSO Scheduling Algorithm

Then adjusting the pheromone on the task with the virtual machine path, which affects the
selection of the ant for the other ant and the next iteration for the ant. In this article, the initial
pheromone, heuristic function and pheromone update rule of ant colony algorithm are improved
by combining the objective function to be optimized.

Experimental results compared to Ant Colony Optimization (ACO) and Min-Min showed the
MO-ACO algorithm satisfies expectation.

II.3. PSO algorithm (particle swarm optimization)

Particle Swarm Optimization (PSO) algorithm, As GA and ACO is one of the most well-
regarded algorithms in the literature of stochastic optimization approaches. It belongs to the
family of swarm-based techniques and is a population-based algorithm. [23]

The PSO algorithm simulates the navigation mechanism of birds in nature. The main inspiration
Is the simple equations proposed by Raynold in 1987 And it was developed by Kennedy and
Eberhart in 1995, in which both researchers got inspired by social and computer sciences [24].

For simulating the interaction of individuals in flying swarms. This swarm of particles
simultaneously explore a problem’s search space with the goal of finding the global optimum
configuration [25]. Oriented to the movements of their neighbors and the best positions in
solution space found so far, particles move in solution space with velocities and randomness.
This plays the role of the explorative mechanism. The orientation to the best so far found
positions is the exploitative counterpart.[23]

In PSO algorithm, candidate solutions (fishes, birds,.etc.) or particles travel through the search
space to find an optimal solution, by interacting and sharing information with neighbor
particles, namely their individual best solution (local best) and computing the neighborhood
best. Also, in each step of the procedure, the global best solution obtained in the entire swarm
is updated. Using all of this information, particles realize the locations of the search space where
success was obtained, and are guided by these successes. In each step of the algorithm, a fitness
function is used to evaluate the particle success.

To model the swarm, in iteration t, each particle n moves in a multidimensional space according
to position (xn[t]) and velocity (vn[t]) values which are highly dependent on local best (x1n[t]),
also known as the cognitive component, and global best (x2n[t]), typically known as the social
component, as follows.

28
Chapter II: The Proposed PSO Scheduling Algorithm

𝑣𝑛 [𝑡 + 1] = 𝜔𝑣𝑛 [𝑡] + ∑ 𝜌𝑖 𝑟𝑖 (𝑥𝑖𝑛 [𝑡] − 𝑥𝑛 [𝑡])

𝑖=1

𝑥𝑛 [𝑡 + 1] = 𝑥𝑛 [𝑡] + 𝑣𝑛 [𝑡 + 1]

Coefficients 𝜔, x1; and x2 assign weights to the inertial influence, the local best, and the global
best when determining the new velocity vn[t+1], respectively. Typically, the inertial influence
is set to a value slightly less than 1.

𝜌1 and 𝜌1 are constant integer values, which represent “cognitive” and “social” components.
However, different results can be obtained by assigning different influences for each
component. For example, some works additionally consider a component called neighborhood
best (which in this case would be a p3).

The parameters r1 and r2 are random vectors with each component generally a uniform random
number between0 and 1. The intent is to multiply a new random component per velocity
dimension, rather than multiplying the same component with each particle’s velocity
dimension. Depending on the application and the characteristics of the problem, tuning all these
parameters properly may lead to better results. However, this is a problem associated with the
branch of parameterized complexity and is far from being completely solved. In fact, Fig II.3
clearly illustrates balls of different sizes.

Fig II.3: Three main causes that may influence a bird’s (particle) trajectory

29
Chapter II: The Proposed PSO Scheduling Algorithm

Where:

a) Red ball is the previous direction it was traveling.

b) Green ball is the best location it personally found thus far.

c) Blue ball is the best location the flock collectively found thus far.

This is an analogy made to the influence of each component from velocity equation namely the
previous velocity of the particle by means of 𝝎 (red ball), its individuality by means of 𝜌1 (green
ball), and how much it cooperates with its society by means of 𝜌2 (blue ball).

In the beginning, particles’ velocities are set to zero and their positions are randomly set within
the boundaries of the search space. The local and global bests are initialized with the worst
possible values, taking into account the nature of the problem. For instance, in a cost problem
where the objective is to minimize the fitness function, particles are initialized with a large
value (tending to infinity). There are a few other parameters that need to be adjusted:

• Population size: Very important to optimize to get overall good solutions inacceptable time
(remember that each particle is a potential solution of the problem).

• Stopping criteria: It can be a predefined number of iterations without getting better results
(known as stagnation), or other criteria depending on the problem (common examples are time
or performance). [24]

II.3.1. PSO Benefits

The main advantage of using the PSO is its simple concept and ability to be implemented in a
few lines of code. Furthermore, PSO also has a short-term memory, which helps the particles
to fly over the local best and the global best positions. Alternatives, such genetic algorithms
(GA), are more complex and, most of the time, they do not consider the previous iteration or
the collective emergent performance. For instance, in GA, if a chromosome is not selected, the
information contained by that individual is lost.[24]

30
Chapter II: The Proposed PSO Scheduling Algorithm

II.4. Modeling job scheduling in Cloud Computing using PSO algorithm

We present a scheduling heuristic optimizing the cost and makespan of job scheduling using
the mapping solution computed by particle swarm optimization-based algorithm. PSO is one of
the latest evolutionary algorithms inspired by the social behavior of fish schooling or bird
flocking. The particle corresponds to an individual bird or fish searching in a natural space.

In the following, I adopt the general model and notation used by existing works on Design and
Implement of a Scheduling Strategy Based on PSO Algorithm [26].

a) Scheduling Problem Formulation

A job scheduling is usually represented by a set of nodes V = {t1, …, tn} represents the tasks in
the workflow applications, and n is the total number of tasks in the scheduling.

Suppose there are a total of m resources in the cloud environment. The resources can be denoted
as R = {R1, … , Rm}. The scheduling problem is to find an optimal mapping M between tasks
and resources according to some optimization objective.

b) the Fitness Function

As mentioned before, makespan is an objective that is critical for this scheduling. In the
following, I formulate several optimization objectives. Let makespantotal(M) denote the
makespan of the job scheduling with respect to the mapping M:

𝒎𝒂𝒌𝒆𝒔𝒑𝒂𝒏𝒕𝒐𝒕𝒂𝒍 (𝑴) = 𝒇𝒊𝒏𝒊𝒔𝒉 𝒕𝒊𝒎𝒆 𝒐𝒇 𝒍𝒂𝒔𝒕 𝒕𝒂𝒔𝒌 − 𝒔𝒕𝒂𝒓𝒕 𝒕𝒊𝒎𝒆 𝒐𝒇 𝒕𝒉𝒆 𝒇𝒊𝒓𝒔𝒕 𝒕𝒂𝒔𝒌 (1)

For the objective of optimizing makespan (such as the work by Zhang et al [13]), the fitness
function can be defined as:

𝒇𝒊𝒕𝒏𝒆𝒔 𝒇𝒖𝒏𝒄𝒕𝒊𝒐𝒏 = 𝑴𝒂𝒌𝒆𝒔𝒑𝒂𝒏𝒕𝒐𝒕𝒂𝒍 (𝑴) (2)

The objective is to minimize Fitness function.

c) Select the Particle Swarm Model

In the PSO algorithm, each particle represents a possible solution. The flock or swarm of
particles is randomly generated initially. Each particle has its own position in the space and a
fitness value, and has the velocity to determine the speed and direction it flies. A group of
candidate solutions (particles) moves around in the search space based on the particles’ updated

31
Chapter II: The Proposed PSO Scheduling Algorithm

position and velocity so that the PSO algorithm can get a optimized solution. Particles in the
search process update themselves by tracking the local best position is the individual best
known position in terms of fitness value reached so far by the particle itself. And Another the
global best position is the best position in the entire population. Suppose the number of particles
is N. The velocity and position of each particle are calculated by formulations (3) and (4).

𝒗𝒌+𝟏
𝒊 = 𝝎𝒗𝒌𝒊 + 𝒄𝟏 𝒓𝟏 (𝒑𝒌𝒊 − 𝒙𝒌𝒊 ) + 𝒄𝟐 𝒓𝟐 (𝒑𝒌𝒈 − 𝒙𝒌𝒊 ) 𝟏≤𝒊≤𝑵 (𝟑)

𝒙𝒌+𝟏
𝒊 = 𝒙𝒌𝒊 + 𝒗𝒌+𝟏
𝒊 𝟏≤𝒊≤𝑵 (𝟒)

𝒑𝒌𝒊 = 𝑩𝒆𝒔𝒕(𝒙𝒌𝒊 , 𝒑𝒌−𝟏

𝒊 ) 𝟏≤𝒊≤𝑵 (𝟓)

𝒑𝒌𝒈 = 𝑩𝒆𝒔𝒕(𝒑𝒌𝟏 , … , 𝒑𝒌−𝟏

𝑵 ) (𝟔)

Where:

𝒗𝒌+𝟏
𝒊 Velocity of particle 𝑖 at iteration 𝑘 + 1

𝒗𝒌𝒊 Velocity of particle 𝑖 at iteration 𝑘

𝒙𝒌+𝟏
𝒊 Position of particle 𝑖 at iteration 𝑘 + 1

𝒙𝒌𝒊 Position of particle 𝑖 at iteration 𝑘

𝒑𝒌𝒊 Best position of particle 𝑖 so far at iteration 𝑘

𝒑𝒌𝒈 Best position of all particle𝑠 so far at iteration 𝑘

𝜔 Inertial weight

c1, c2 Acceleration coefficients (positive constants)

𝑟1, 𝑟2 Random numbers in [0,1]

Particle’s best position is calculated by the fitness function through (5) and (6). The velocity is
calculated by three factors each iteration. They are current velocity, local best position and
global best position. The position is updated according to its current position and updated

32
Chapter II: The Proposed PSO Scheduling Algorithm

velocity. These ensure the particles search around the local and global best positions and
converge to a global best position in the limited iteration.

In the job scheduling problem, the particle represents a mapping between m resource and n task.
Each particle position represents one schedule, and each position variable represents the job
assigned to the corresponding resources. In Fig II.3, 11 possible particles for the mapping
between 8 resources and 11 tasks is illustrated. The evaluation of each particle is performed
based on fitness function.

T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T11

R4 R2 R4 R5 R1 R7 R5 R3 R6 R8 R1

Fig II.4: A sample particals

d) the Termination Criteria of the Algorithm

The most commonly used termination criteria of PSO algorithm are preset a maximum number
of iterations, or terminate algorithm when the fitness value of solutions have no significant
changes after multiple iterations in the searching process. In our case we will choose the first
termination criteria which is the number of iterations.

e) The pseudo code for PSO

Tab II.2 lists the steps of PSO algorithm. The initial step is to initialize the swarm, set randomly
for each particle’s his position and velocity. If iteration criterion is not met, the algorithm
repeatedly does the following: firstly calculate each particle’s fitness value by using the fitness
function given in (2), and then update its local best position using (5); calculate global best
position using (6); for each particle, update its velocity and position using (3) and (4). Finally,
the global best position is the optimal mapping solution.

33
Chapter II: The Proposed PSO Scheduling Algorithm

1: Initialize particles’ position and velocity randomly.

2: While stopping criterion is not satisfied do

3: For each particle do

4: Calculate its fitness value using fitness function.

5: Update its local best position

6: End For

7: Update global best position.

8: For each particle do

9: Update its velocity and position.

10: End For

11: End While

12: Return global best position.

Tab II.2: The pseudo code of PSO algorithm

34
Chapter II: The Proposed PSO Scheduling Algorithm

Begin

Initialization: Initialize the swarm (size n, particle velocity, particle

position, inertia weight, and acceleration constants c1 and c2)

Calculate the fitness function for each particle and determine pbest of
each particle and gbest of swarm.

Update the velocities of all particles According to

No equations (3)

Update the position of each particle According to

equations (4)

Stopping criteria
satisfied?

Yes

Find the maximum completion time of all jobs

(makespan)

End

Fig II.5: The flow chart of PSO for job scheduling in cloud computing

35
Chapter II: The Proposed PSO Scheduling Algorithm

II.5. Conclusion

Cloud computing is an emerging paradigm in the computing world, in which tasks scheduling
is one of the most important areas of research. An efficient algorithm-based job scheduling is
necessary to achieve user satisfaction and maximize profits for cloud computing service
providers.

In this chapter on an attempt to show the recent work linked to our project and which will make
the fundamental aspects as well as the main challenges that we will be tempted to materialize
in the third chapter by relying on the CloudSim library and the language of java programming
in order to achieve to implement the Particles swarm optimization algorithm to try to optimize
the scalable partition of distributed computation.

36
 I.
II.
III. Chapter III

V. Implementation and
Experiment Result
Chapter III: Implementation and Experiment Result

III.1. Overview

Tasks scheduling is one of the keys for large-scale Cloud applications. Extensive research on
all issues in real environment is extremely difficult because it requires developers to consider
network infrastructure and the environment, which may be beyond the control. In addition, the
network conditions cannot be controlled or predicted. Performance evaluations of workload
models and Cloud provisioning algorithms in a repeatable manner under different
configurations are difficult. Therefore, simulators are developed. To understand and apply
better the state-of-the-art of cloud computing simulators. Many simulators are available for the
modeling the cloud environment like CloudSim, GreenCloud, CloudAnalyst, etc.

This chapter explains the computational experiments for evaluating the performance of the
particle swarm optimization (PSO) with SJF-PSO using the CloudSim simulator.

III.2. Cloud computing simulation

Cloud Computing is faced many problems by many Industries which are directly or indirectly
related to the Information Technology. This field is growing very fastly as many of the big
players in the Information Technology field like Microsoft, Google, Amazon, SAP are investing
a lot of money to get improved results. They feel that it is the future of technology.

There are variety of simulator tools for modelling and simulation of large-scale Cloud
computing environments (Fig III.1). Generally, we can designate between two types of
simulators: graphical user interface (GUI) simulators or programming language-based
simulators (like Java for example).[29]

Fig III.1: Cloud Computing Simulation frameworks [29]

37
Chapter III: Implementation and Experiment Result

In this work I chose the CloudSim Simulator because of my participation in on-line certified
training course on “learn basics of CloudSim” from the Udemy site [33], by the trainer
Anupinder singh, to simulate the proposed algorithms for job Scheduling in Cloud Computing
environment. I will discuss the details of this simulator in the next section.

III.2.1. Reasons for adopting for modeling and simulation [30]

• Various cloud computing data centers
• Virtualization of server hosts, associated with customized policies for provisioning host
resources to VMs
• Specific purpose software containers
• Power-aware computational resources
• Different datacenter network topologies as well as message-passing applications
• Dynamic addition or removal of simulation components
• Simulation can be stop and resume
• User-defined policies provide support for allotment of hosts to VMs and policies for
allotment of host resources to VMs
III.3. Software Tools Used in the Research
III.3.1. Eclipse

Eclipse is an integrated development environment (IDE) used in computer programming. It

contains a base workspace and an extensible plug-in system for customizing the environment.
Eclipse is written mostly in Java and its primary use is for developing Java applications, but it
may also be used to develop applications in other programming languages via plug-ins,
including Ada, ABAP, C, C++, C#, Clojure, COBOL, D, Erlang, Fortran, Groovy, Haskell, Ja
vaScript, Julia, Lasso, Lua, NATURAL, Perl, PHP, Prolog, Python, R, Ruby (including Ruby
on Rails framework), Rust, Scala, and Scheme. It can also be used to develop documents
with LaTeX (via a TeXlipse plug-in) and packages for the software Mathematica. Development
environments include the Eclipse Java development tools (JDT) for Java and Scala, Eclipse
CDT for C/C++, and Eclipse PDT for PHP, among others.[32]

The initial codebase originated from IBM VisualAge. The Eclipse software development
kit (SDK), which includes the Java development tools, is meant for Java developers. Users can
extend its abilities by installing plug-ins written for the Eclipse Platform, such as development
toolkits for other programming languages, and can write and contribute their own plug-in
modules.[32]

38
Chapter III: Implementation and Experiment Result

III.3.2. CloudSim

CloudSim is a simulation tool that developed in CLOUDS laboratory in University of

Melbourne for cloud computing environment. It enables users to evaluate specific system issues
without taking into consider the low-level facts associated to cloud-based infrastructures and
services. It means support seamless modeling, simulation and implementing on designing cloud
computing framework. CloudSim is an autonomous platform that supports of large-scaled cloud
platform for describing users, applications, datacenters, brokers, scheduling and provisioning
policies. [30]

CloudSim is an open source software that enables users to test application in controlled and
repeatable environment, find the system bottlenecks without the need of real clouds and try
different configurations for developing adaptive provisioning techniques.[30]

III.3.2.1. The pros of CloudSim Simulation [30]

• The elasticity of significant configurations
• Support for diverse cloud Environnent
• It enables modeling of application services in any environment
• cloud based application implementation in minimum time and minimum effort
• Ease of use and customization
III.3.2.2. The cons of CloudSim simulation [30]
• Does not support Graphical User Interface (GUI) to generate results
• Support restricted workload transfer generator due to basic network model
• Not helpful for modeling for parallel experiments;
• Difficult to analyzing simulation results
• So researcher might need to analyze data using other tools such as Excel.
III.3.2.3. CloudSim Versions [30]

There are different versions of CloudSim available for cloud computing environment. They are

a) Version 1.0:

Version 1.0 was the first version of CloudSim, and it was released on April 7, 2009. It supports
seamless scale modeling and simulation, moreover implementing on designing cloud
computing framework. CloudSim is an autonomous platform that supports of large scaled cloud
platform for describing users, applications, datacenters, brokers, scheduling and provisioning
policies.

39
Chapter III: Implementation and Experiment Result

b) Version 2.0:

It was released on May 27, 2010. This version improved the simulation that allowed superior
scalability and efficiency of simulation and addition and removal of simulation components
during simulation execution.

c) Version 2.1

It was released on July 27, 2010. It has the capability to migrate using apache Maven. Maven
provides various tools and plug-in for simplifies the project such as bug fixes, refactoring and
removal of obsolete code.

d) Version 2.1.1:

It was released on Feb 1, 2011. It has capability to fix some bug which comes in version 2.1.

e) Version 3.0:

It was released on June 11, 2012. This version of CloudSim also updates bug fixes. It also
updates in this are new VM scheduler, novel datacenter network model, new VM allocation and
selection policies, new power models, new workload tracks, support for external workloads and
support for user define end of simulation. It support elimination of a few classes have been
made like Cloud coordinator, PowerPe, Sensor and Power.PeList. This version has capability
to fix some API changes and bugs.

f) Version 4.0:

This is the latest version of CloudSim was released Jan, 2017. New applications of Cloud
computing are emerging such as Internet of Things and Big Data, and new technologies are
being incorporated to the fabric of Cloud data centers such as Container virtualization. It
support for container virtualization and lots of bug fixes.

40
Chapter III: Implementation and Experiment Result

III.3.2.4. Architecture of CloudSim Simulation

Fig III.2: Layered CloudSim architecture. [31]

The above diagram demonstrates the layered architecture of CloudSim Simulation Toolkit.
The CloudSim Core simulation engine provides support for modeling and simulation of
virtualized Cloud-based data center environments including queuing and processing of events,
creation of cloud system entities (like data center, host, virtual machines, brokers, services, etc.)
communication between components and management of the simulation clock. [31]

The CloudSim layer provides dedicated management interfaces for Virtual Machines, memory,
storage, and bandwidth. Also, it manages the other fundamental issues, such as provisioning of
hosts to Virtual Machines, managing application execution, and monitoring dynamic system
state (e.g. Network topology, sensors, storage characteristics, etc), etc. [31]

The User Code layer is a custom layer where the user writes their own code to redefine the
characteristics of the stimulating environment as per their new research findings. [31]

41
Chapter III: Implementation and Experiment Result

the fundamental classes of CloudSim, which are also the building blocks of the simulator. The
overall Class design diagram for CloudSim is shown in Fig III.3

Fig III.3: CloudSim class design diagram. [31]

a) Cloudlet

This class models the Cloud-based application services (such as content delivery, social
networking, and business workflow). CloudSim orchestrates the complexity of an application
in terms of its computational requirements. Every application service has a pre-assigned
instruction length and data transfer (both pre and post fetches) overhead that it needs to
undertake during its life cycle. This class can also be extended to support modeling of other
performance and composition metrics for applications such as transactions in database-oriented
applications. [31]

b) Datacenter

This class models the core infrastructure-level services (hardware) that are offered by Cloud
providers (Amazon, Azure, App Engine). It encapsulates a set of compute hosts that can either
be homogeneous or heterogeneous with respect to their hardware configurations (memory,
cores, capacity, and storage). Furthermore, every Datacenter component instantiates a
generalized application provisioning component that implements a set of policies for allocating
bandwidth, memory, and storage devices to hosts and VMs. [31]

42
Chapter III: Implementation and Experiment Result

c) DatacenterBroker

This class models a broker, which is responsible for mediating negotiations between SaaS and
Cloud providers; and such negotiations are driven by QoS requirements. The broker acts on
behalf of SaaS providers. It discovers suitable Cloud service providers by querying the CIS and
undertakes online negotiations for allocation of resources/services that can meet the
application’s QoS needs. Researchers and system developers must extend this class for
evaluating and testing custom brokering policies. The difference between the broker and the
CloudCoordinator is that the former represents the customer (i.e. decisions of these components
are made in order to increase user-related performance metrics), whereas the latter acts on
behalf of the data center, i.e. it tries to maximize the overall performance of the data center,
without considering the needs of specific customers. [31]

III.3.2.5. Basic scenario of CloudSim Simulation

The main scenario of CloudSim simulation is presented in figure III.3. This scenario may be
described the detail of each component of CloudSim as follows: that Data centers (DC) provides
resource such as more than one host. Host is physical machine which allocates more than one
VMs. Virtual machine (VM) are logical equipment on that the cloudlet will be executed. Broker
has DC characteristics that allow it to submit virtual machines to the exact host. Cloud
Information Service (CIS) is responsible for the listing of resources, indexing, as well as
discovering the efficiency of data centers [30].

Fig III.3: Basic scenario of Simulation.[30]

43
Chapter III: Implementation and Experiment Result

CloudSim gives classes for virtual machines, Datacenters, DatacenterBroker, cloudlets and
scheduling methods. Different stages of CloudSim lifecycle is shown in Fig III.4.

Begin

Initialize the CloudSim package

Creation of Datacenters

Creation of Broker

Creation of VMs

Creation of Cloudlets

Start Simulation

Stop Simulation

Output the Results

End

Fig III.4: CloudSim lifecycle.

III.4. Implementation and Experiment Result

The implementation code of the proposed job scheduling algorithm PSO has been written by

using Eclipse Java 2019-09 this code is implemented to extract results that used to evaluate
the performance of the proposed job scheduling algorithm. The computer system specification
used to implement the algorithm is:

• Processor: Intel® core™ i5-8350U CPU @ 1.70GHz (8 CPUs), 1.9GHz

• Installed memory (RAM): 16384MB
• Hard disk drive: 500 GB (SSD)
• Operating System: Win10 Professional 64bits

44
Chapter III: Implementation and Experiment Result

I use the JSwarm[17] package to conduct the simulation in PSO. The PSO class, shown in
Fig III.1, aggregates Swarm class that also has Particle and FitnessFunction class. Swarm,
Particle and FitnessFunction are in JSwarm package. Then I define our own SchedulerParticle
class for PSO scheduling algorithm that extends the Particle class. For FitnessFunction class, I
extend it to my fitness function class, which is SchedulerFitnessFunction.

PSO
1
1

Swarm
1 1

1 1

Particle FitnessFunction

schedulerParticle schedulerFitnessFunction

Fig III.5: Class diagram of PSO scheduling

In the experiment, I test two algorithms. PSO algorithm and SJF-PSO algorithm are the
scheduling algorithms using the Makespan fitness function. The SJF-PSO proposed algorithm

is PSO algorithm added with sorting part.

SJF or the Shortest Job First is One of the proposed tasks scheduling algorithms. The task in
this algorithm has the lowest execution time, which its process having the smallest CPU burst
gets a high priority and executes at the first by the CPU. The remaining tasks wait for the
execution during this time, and finally the task that has the highest execution time will be
executed at the end. The criteria of Shortest Job First (SJF) scheduling algorithm is that it
process the smallest task first.[28]

In SJF-PSO, I have modified the CloudletSubmition method under the DataCenterBroker class
Where I Calculate the cloudlets length for all cloudlets and submit them in an ascending order

I generate a text file filled with random integers, these random integers are a lengths for more
than 1000 of cloudlets.

45
Chapter III: Implementation and Experiment Result

I write a class (InfBase) for the different basic information need in the scheduling simulation,
to make it easier to control the number of Cloudlets, number of VMs, number of Datacenters,
Number of Particles, and the Number of Iterations.

Fig III.6: The InfBase Class

46
Chapter III: Implementation and Experiment Result

Tab III.1 shown the different Simulation Parameters used in my implementation.

Parameter Values
Number of Data-Centers 1
System Architecture X86
Number of Hosts 4
VMM XEN
OS Linux
Transfer Rate 15 MB/s
Numbers of VMs 8
CPU (PEs Number) 4
RAM per VM 512 MB
Bandwidth 1000
PE‟s MIPS per VM 1000
Image Size 20000
Policy Type Time Shared
Tab III.1: Simulation Parameters.

The data center that I use in the simulation (shown in Fig III. 9) consists of four hosts with four
processing element (PEs) and each host having two VMs. They are constructed by cloud-based
interface provided by CloudSim.

Datacenter1

host0 host1 host2 host3

VM0 VM1 VM2 VM3

VM4 VM5 VM6 VM7

Fig III.7 Experimental datacenter infrastructure

47
Chapter III: Implementation and Experiment Result

CloudSim life cycle starts with initialization of CloudSim environment and ends with
simulation results. Scheduling algorithm is applied after creation of cloudlets. The following
table shown the pseudo code for the simulation steps:

Step 1: Initialize the CloudSim package by calling CloudSim.init(int numUser, Calendar cal,

boolean traceFlag);

Step 2: Create Datacenters and set the needed Datacenter Characteristics;

Step 3: Create Broker;

Step 4: Create VMs and set the configuration of each VM. Add VMs to the VM list. Submit the

VM list to the broker;

Step 5: Create Cloudlet. Similarly, set the properties of each Cloudlet and add it into the cloudlet

list. Submit the cloudlet list to the broker;

Step 6: Implement a PSO tasks scheduling function in the DatacenterBroker;

Step 7: Bind the cloudlets to VMs by calling the function defined in DatacenterBroker to apply

your own task scheduling algorithm;

Step 8: Start the simulation by calling CloudSim.startSimulation ();

Step 9: Get the results by printing them out on the console;

Step 10: Stop the simulation by calling CloudSim.stopSimulation ();

Tab III.2: pseudo code for the job scheduling simulation steps.

48
Chapter III: Implementation and Experiment Result

Once the configuration is done, we launch the simulation through the Button "Run". The
simulation results presented by the following windows, including the optimal solution given by
our "best fitness value" application, then the assignment of cloudlets to VMs after the creation
of all the necessary entities (data center, broker, etc.).

Fig III.8 Result of the Simulation shown the best makespane.

49
Chapter III: Implementation and Experiment Result

Fig III.9: creation of cloud entities.

Fig III.10: execution statistics for 10 cloudlets.

50
Chapter III: Implementation and Experiment Result

The result of execution statistics for 20 cloudlets with the proposed SJF-PSO, is shown in the
Fig III.11. The cloudlets 8,12,3 are executed by the VM ID=0 starting by the lowest length.

Fig III.11: execution statistics for 20 cloudlets with the proposed SJF-PSO.

51
Chapter III: Implementation and Experiment Result

For 20 cloudlets and 8 particles The following table shows the result of the best fitness value
(Makespan) with respect to the number iterations:

No Of Iteration Makespan PSO Makespan SJF-PSO

3 11411,95622 10859.310803448961

6 11159,65632 9264.911006568873

9 11520,94829 9259.853203613835

12 9707,843941 9700.193965505692
VM6 VM7
15 8854,685946 9555.861414496841

18 9613,177707 7794.950423544414

Tab III.3: the result of the Makespane with respect to the number of iterations for PSO and
SJF-PSO

For 10 Iteration and 8 Particles the following table shows the result of the best fitness value
with respect to the number Cloudlets

No Of Cloudlets Makespan PSO Makespan SJF-PSO

100 6343,59362992529 5445,56152360159

200 10051,30646791640 10976,86198913910

300 15172,60993807690 16048,74864530280

400 20981,96963226440 21016,13944651330

VM6 VM7
500 22847,79779719150 23308,57764281200

600 31176,72352599040 34001,87216986880

Tab III.4: the result of the Makespane with respect to the number of Cloudlets for PSO and
SJF-PSO

52
Chapter III: Implementation and Experiment Result

The simulation results obtained by the solution clearly demonstrate an efficient allocation using
the SJF-PSO which allows a considerable optimization of the makespan of all the cloudlets at
the cloud level compared to the result given by the PSO, the following two images show the
results obtained.

35000
Best fitness value (Makespane)

30000

25000

20000

15000 Makespane PSO

Makespane SJFPSO
10000

5000

0
10 20 30 400 500 600
Number of Cloudlet

Fig III.11: Makespan comparison between PSO and SJF-PSO with respect of number of
Cloudlets.

14000
the best fitness value (Makespane)

12000

10000

8000

6000 Makespane PSO

Makespane SJF-PSO
4000

2000

0
3 6 9 12 15 18
Number of Iteration

Fig III.12: Makespan comparison between PSO and SJF-PSO with respect of number of
Iteration.

53
Chapter III: Implementation and Experiment Result

III.5. Conclusion

CloudSim Simulation Toolkit is one of the most used simulators for the implementation of the
cloud-related research problem. The training I did [33], helped me to follow the simulation-
based approach of CloudSim and can leverage various benefits like, Testing of services and
scenarios in a controllable environment of cloudSim without spending a money on the real
cloud. Also simulating the small- or large-scale infrastructure to evaluate different sets of
proposed algorithms as well as workload along with the resource performance. Ultimately
facilitating the development, testing, and deployment of adaptive algorithms for resource
provisioning techniques. And finally Optimizing the performance of various research-based
algorithms with the cloudSim simulation engine before writing a research paper.

In this chapter, we have proposed our solution in the form of a java application which presents
a simulation of a Cloud Computing in integration with the CloudSim library, of which we are
interested in job scheduling on our Cloud by the implementation of two algorithm which are
PSO and SJF-PSO
The proposed PSO scheduling considers minimizing the makespan value of Job scheduling that
cannot obtain the lowest Makespan. Therefore, I propose the SJF-PSO algorithm which firstly
start by sorting the cloudlets in ascending order then we apply the same first PSO algorithm.
The experimental results show that by adding sorting part into the PSO algorithm the Makespan
can be reduced even further.

54
 I.
II.
III. Chapter III

IV. Conclusion
Conclusion

Conclusion

Cloud computing is a distributed based computer paradigm which is used by users to get good

quality with less cost. As task scheduling is a challenge in cloud computing, different algorithms

have been suggested and applied to get better results regarding utilization of system resources,

response time and satisfaction of user demands. In this study, two proposed algorithms for the

task in cloud namely PSO, and SJF-PSO. These algorithms are implemented within the

environment of CloudSim, in order to compare them.

The results show that when using SJF-PSO to scheduling number of Cloudlets, the makespane

becomes less than when using PSO, using the same numbers of cloudlets and the Cloud

Computing environments. For future work, it is worth to investigate the effect of other

parameters such as VMs, datacenters, memory, bandwidth for network and storage in cloud

environments, and reflect that in real physical environment.

55
References

Articles, E-books:

[1] Peter Mell and Timothy Grance, The NIST definition of cloud computing, 2011,

p (2-3).

[2] Yuri Demchenko, Marc X. Makkes, Rudolf Strijkers, Rudolf Strijkers, Intercloud

Architecture for Interoperability and Integration,The 2nd NetCloud2012 Workshop on Network

Infrastructure Services as part of Cloud Computing, in Proc. The 4th IEEE Conf. on Cloud

Computing Technologies and Science, 2012, p (3 – 6)

[3] Keyun Ruan, Joe Carthy, Cloud Computing Reference Architecture and Its Forensic

Implications: A Preliminary Analysis, Center for Cybersecurity and Cybercrime Investigation,

University College Dublin, 2013, p (3-8)

[4] K. Chandrasekaran, Essentials of CLOUD COMPUTING, CRC PressTaylor & Francis

Group, 2015, p (38-39)

[5] The Open Group Guide, Cloud Computing for Business,2010, p (12)

[6] Bernd Gastermanna, Markus Stoppera , Anja Kossikb , Branko Katalinicc, Secure

Implementation of an On-Premises Cloud Storage Service for Small and Medium-Sized

Enterprises, 25th DAAAM International Symposium on Intelligent Manufacturing and

Automation, Elsevier , 2014, p (576)

[7] Chunye Gong, Jie Liu, Qiang Zhang, Haitao Chen and Zhenghu Gong, The

Characteristics of Cloud Computing, 2010, In 39th International Conference on Parallel

Processing Workshops, IEEE, 2010, p (275–279).

[8] Professional Xen Virtualization, Wiley, 2008, p (12).

[9] Massimo Cafaro, Giovanni Aloisio, Grids, clouds, and virtualization, Springer London,

2011, p (16)

56
References

[10] James E Smith, Ravi Nair, The architecture of virtual machines, the IEEE Computer

Society, 2005, p (32).

[11] Michael Nelson, Beng-Hong Lim, and Greg Hutchins, Fast transparent migration for

virtual machines. In Proceedings of the Annual Conference on USENIX Annual Technical

Conference, USENIX Association, 2005, p (1-4)

[12] Christopher Clark, Keir Fraser, Steven Hand, Jacob Gorm Hansen†, Eric Jul†, Christian

Limpach, Ian Pratt, Andrew Warfield, Live migration of virtual machines. In Proceedings of

the 2Nd Conference on Symposium on Networked Systems Design & Implementation Volume

2, USENIX Association, 2005, NSDI'05, p (273).

[13] Md. Abdur Rashid, Data Center Architecture Overview Article, National Academy for

Planning and Development (NAPD), Ministry of Planning, Bangladesh, 2019, p(3)

[14] Gary Lee, Cloud Networking: Understanding Cloud-based Data Center Networks,

Elsevier Inc. 2014, p (4)

[15] Linz Tom, V. R. Bindu, Task Scheduling Algorithms in Cloud Computing: A Survey,

Inventive Computation Technologies, Springer Nature Switzerland AG, 2020, p (360)

[16] Kai Hwang, Geoffrey Fox, Jack Dongarra, Distributed Computing: Clusters, Grids and

Clouds, Elsevier, 2012, p (221-222)

[17] Patricia Takako Endo, Glauco Estácio Gonçalves, Daniel Rosendo, Demis Gomes, Guto

Leoni Santos, André Luis Cavalcanti Moreira, Judith Kelner, Djamel Sadok, Mozhgan

Mahloo,Highly Available Clouds: System Modeling, Evaluations, and Open Challenges,

Research Advances in Cloud Computing Sanjay Chaudhary Gaurav Somani Rajkumar Buyya

, Springer Nature Singapore, 2017, p (47)

57
References

[18] Sung-Soo Kim, Ji-Hwan Byeon, Hong Yu, Hongbo Liu, Biogeography-based

optimization for optimal job scheduling in cloud computing, Applied Mathematics and

Computation, Elsevier Inc, 2014, p (266-280)

[19] Safwat A. Hamad, Fatma A. Omara, Genetic-Based Task Scheduling Algorithm in

Cloud Computing Environment , (IJACSA) International Journal of Advanced Computer

Science and Applications, Vol. 7, 2016, p ( 550-556)

[20] Ali Belgacem, Kadda Beghdad-Bey, Hassina Nacer, Task Scheduling in Cloud

Computing Environment: A Comprehensive Analysis, Springer Nature Switzerland AG 2019,

p (16-17)

[21] Raja Manish Singh, Sanchita Paul, Abhishek Kumar, Task Scheduling in Cloud

Computing: Review, (IJCSIT) International Journal of Computer Science and Information

Technologies, Vol. 5 (6), 2014, p (1)

[22] Gurtej Singh, Amritpal Kaur, Bio Inspired Algorithms: An Efficient Approach for

Resource Scheduling in Cloud Computing, International Journal of Computer Applications

Volume 116 – No. 10, April 2015, p (16)

[23] Seyedali Mirjalili, Evolutionary Algorithms and Neural Networks Theory and

Applications, Studies in Computational Intelligence Volume 780, Springer Nature 2019, p15

[24] Micael Couceiro, Pedram Ghamisi, Fractional Order Darwinian Particle Swarm

Optimization Applications and Evaluation of an Evolutionary Algorithm, Springer Cham

Heidelberg New York Dordrecht London, 2016, p (1-3)

[25] Bastien Chopard, Marco Tomassini An Introduction to Metaheuristics for Optimization,

Springer Nature Switzerland AG, 2018, p (97)

58
References

[26] Suqin Liu, Jing Wang, Xingsheng Li, Jun Shuo, and Huihui Liu, Design and Implement

of a Scheduling Strategy Based on PSO Algorithm, Advances in Swarm Intelligence First

International Conference ICSI 2010, Part II, 2010, p (535-538)

[27] Qiang Guoa, Task scheduling based on ant colony, optimization in cloud environment

2017 5th International Conference on Computer-Aided Design, Manufacturing, Modeling and

Simulation CDMMS, 2017, p (1-11).

[28] Mokhtar A. Alworafi Atyaf Dhari Asma A. Al-Hashmi A. Basit Darem Suresha , An

Improved SJF Scheduling Algorithm in Cloud Computing Environment, International

Conference on Electrical, Electronics, Communication, Computer and Optimization

Techniques (ICEECCOT), IEEE, 2016, p (208)

[29] Riya Joshi1 , Study and Comparison of VM Scheduling Algorithm in Cloud Computing

Using CloudSim Simulator, International Journal for Research in Applied Science &

Engineering Technology (IJRASET) Volume , 2018, p (1751).

[30] Mohammad Oqail Ahmad and Rafiqul Zaman Khan, Cloud Computing Modeling and

Simulation using CloudSim Environment, International Journal of Recent Technology and

Engineering (IJRTE), Volume-8 Issue-2, 2019, p (5439).

[31] Rodrigo N. Calheiros1, Rajiv Ranjan2, Anton Beloglazov1, C´esar A. F. De Rose3 and

Rajkumar Buyya1, CloudSim: A Toolkit for Modeling and Simulation of Cloud Computing

Environments and Evaluation of Resource Provisioning Algorithms, Software: Practice and

Experience, Wiley Press, 2011. P (30).

Web site:

[32] https://en.wikipedia.org/wiki/Eclipse_(software).

59
References

Certified Training Course :

[33] https://www.udemy.com/course/learn-basics-of-cloudsim/

60
 Abstract
Cloud computing provides a pool of virtualized computing resources and adopts pay-per-use
model. Task Scheduler is one of the most important cloud computing problems. And that cannot
be scheduled using single criteria, but according too many of the criteria's and rules. These rules
and criteria's to be agreed upon by the service provider and the user. In this report, I present a
particle swarm optimization (PSO) based scheduling algorithm to minimize the completion
execution time makespan. Experiment is conducted by varying computation of tasks, number
of particles and makespan in fitness function. Then I modify the proposed PSO by adding a
method for sorting the tasks by them length to execute them by shortest job first (SJF). The two
proposed algorithm are implemented by using the CloudSim simulator. By comparing the two
proposed algorithms (PSO and SJF-PSO) the result given by (SJF-PSO) algorithm is the lower
Makespan.
Keywords:
Cloud Computing, scheduling algorithm, Particle Swarm Optimization (PSO), Shortest
Job First (SJF), Makespan, CloudSim.

Résumé
Le cloud computing fournit un pool de ressources informatiques virtualisées et adopte un
modèle de paiement à l'utilisation. L’ordonnancement de tâches est l'un des problèmes de cloud
computing les plus importants. Et cela ne peut pas être planifié en utilisant des critères uniques,
mais en fonction d'un trop grand nombre de critères et de règles. Ces règles et critères doivent
être convenus par le fournisseur de services et l'utilisateur. Dans ce rapport, je présente un
algorithme d’ordonnancement basé sur l'optimisation des essaims de particules (PSO) pour
minimiser le temps d'exécution. L'expérience est menée en faisant varier le calcul des tâches,
le nombre de particules et le makespan dans la fonction de fitness. Ensuite, je modifie le PSO
proposé en ajoutant une méthode pour trier les tâches par leur longueur afin les exécuter par la
tâches la plus courte en premier (SJF). Les deux algorithmes proposés sont implémentés en
utilisant le simulateur CloudSim. En comparant les deux algorithmes proposés (PSO et SJF-
PSO), le résultat donné par l'algorithme (SJF-PSO) donne le temps d’exécution inférieur.
Mots clés:
Cloud Computing, algorithme d’ordonnancement, L'optimisation par essaims
particulaires (PSO), plus court processus en premier (SJF),temps d’exécution, CloudSim.
 ‫الخالصة‬
‫توفر الحوسبة السحابية مجموعة من موارد الحوسبة االفتراضية وتعتمد على نموذج دفع لكل استخدام‪ .‬يعد برنامج جدولة‬
‫المهام أحد أهم مشكالت الحوسبة السحابية‪ .‬وال يمكن جدولة ذلك باستخدام معيار واحد‪ ،‬ولكن وفقًا لعدد كبير جدًا من المعايير‬
‫والقواعد‪ .‬يتم االتفاق على هذه القواعد والمعايير من قبل مزود الخدمة والمستخدم‪ .‬في هذا التقرير‪ ،‬أقدم خوارزمية جدولة‬
‫تعتمد على خوارزمية حركة الجزيئات)‪ ، (PSO‬لتقليل وقت تنفيذ‪ .‬يتم إجراء التجربة عن طريق حساب المهام المتفاوتة‪،‬‬
‫وعدد الجسيمات واالمتداد في وظيفة اللياقة البدنية‪ .‬ثم أقوم بتعديل ‪ PSO‬المقترح عن طريق إضافة طريقة لفرز المهام حسب‬
‫طولها لتنفيذها بأقصر مهمة أوالً )‪ (SJF‬يتم تنفيذ الخوارزميتين المقترحتين باستخدام محاكي ‪ CloudSim‬بمقارنة‬
‫الخوارزميتين المقترحتين (‪ ، ) SJF-PSO،PSO‬حيث قدمت خوارزمية )‪ (SJF-PSO‬أدنى وقت في تنفيذ المهام‪.‬‬
‫الكلمات الدالة‪:‬‬
‫الحوسبة السحابية ‪ ،‬خوارزمية الجدولة ‪ ،‬تحسين حركة الجسيمات )‪ ، (PSO‬أقصر مهمة أوالً )‪ ، (SJF‬وقت التنفيد‬
‫‪CloudSim ،‬‬