Вы находитесь на странице: 1из 5

INTERNATIONAL JOURNAL OF TECHNOLOGICAL EXPLORATION AND LEARNING (IJTEL)

www.ijtel.org


IJTEL, ISSN: 2319-2135, VOL.3, NO.3, JUNE 2014 479

Hierarchical Organization of Data Centre to
Improve Quality of Services (QoS)
Pooja Devi
Student at DCRUST, Murthal
Gurgaon, India
Kavita Rathi
Assistant Professor at DCRUST, Murthal
Sonepat, India

Abstract The cloud computing has become the buzz topic
in today IT industry because it provides services at lower
cost and high availability of services and supports the
sharing of hardware and software simultaneously with
multiple users. This paper exploits the coordination
between data centre (DC) and user base (UB) to enhance
the performance of data centre by determining the
position of data centre relative to user base. Computing
architecture is based on weak client (user base) and
powerful data centre.
Keywords- Cloud Analyst; Cloud computing; Cloudlet;
resource- rich mobile computing.
I. INTRODUCTION
Cloud computing is getting advanced day by day, and now
it has gain popularity in information services. Many companies
such as Microsoft, Amazon, Red Hat, and IBM are investing
cloud computing infrastructure and research. Cloud computing
is a synthesis of computing services like software as a service
(SaaS), platform as a service (PaaS), infrastructure as a service
(IaaS) [2]. The purpose of cloud computing is to provide
reliable, fault- tolerant, secure and scalable infrastructure for
hosting internet based application services. On one hand there
are cloud service providers who provide and manage the large
scale computing infrastructure at a cheaper price and
eliminating the higher cost of setting up application
deployment environment and provide services in a very
flexible manner. On the other hand many social networking
sites and e-commerce application minimize cost and improve
quality of services to the end users by taking benefits of such
services. But bringing these two ends together there are several
factors that will impact the net benefits such as distribution of
user base, the available internet infrastructure, dynamic nature
of user base etc[5].
Architecture of cloud computing is composed of resource
rich data centre facilities and resource poor clients
background. Researchers and developers have analysed that we
need to test the cloud environment before real time
implementation but it is not easy to study the impact of several
geographical factors. Cloud Analyst is a simulation technique
to investigate such type of behaviour [5].
Another alternative is to use Cloud-Sim framework [4].
CloudSim can be use to configure number of user bases, data
centres, scheduling and allocation policies. By using Cloud-
Sim researchers and developers can test the newly developed
application service in a controlled and easy to set up
environment.
II. CLOUD ANALYST
Cloud computing and distributed system laboratory at
Australia developed cloud analyst simulator with cloud sim
functionality [1]. On the top of Grid- Sim Cloud- Sim is
developed and on the top of Cloud-Sim Cloud Analyst is
developed.
A. New Features
Some new features have been introduced in Cloud Analyst
[3].
Application users: These are the autonomous entities
that need to configure.
Internet: It is used for data transmission across internet.
Simulation defined by time period: In Cloud-Sim
process takes place based on some predefine events.
There is a need to generate events until the set time
period expires.
Service brokers: Management of the routing of user
requests based on the service broker polices like
service proximity based routing, performance
optimized routing, dynamically reconfiguring router.
GUI and ability to save simulation results: Cloud-Sim
is GUI, it makes user to do configuration the
simulation easy. User can save the simulation results in
the form of pdf file for future use.
B. Components of Cloud Analyst [1]
Region: In cloud Analyst the world is divided into 6
Regions.
User Bases: User base may be a single user or group of
hundreds or thousands of users that are used to
generate traffic and access cloud services.
Data Centre: Computing services are provided to user
bases via data centre which process the user requests.
If more than one data centre is available then service
broker algorithm (closest data centre, optimum
INTERNATIONAL JOURNAL OF TECHNOLOGICAL EXPLORATION AND LEARNING (IJTEL)
www.ijtel.org


IJTEL, ISSN: 2319-2135, VOL.3, NO.3, JUNE 2014 480

response time and dynamically reconfigure router) is
used to decide which data centre is used to process user
request.
VmloadBalancer: User generated request is first send
to VmloadBalancer which is responsible to allocate the
load based on load balancing policy (round robin,
throttled, active monitoring load balancer) to the
available VM.
Internet Cloudlet: It is a group of user requests
containing information about size of request, size of
input output files, originator and target application id.
III. RELATED WORK
N. Bhargava et.al [1] in Performance Analysis of Cloud
Computing for Distributed Client by using cloud analyst
analyse the data transfer cost, virtual machine cost and finally
calculate the grand total cost and conclude that performance of
data centre is effected by distance and number of user.
Zia, Khan et.al [2] in Scheme to Reduce Response Time in
Cloud Computing Environment explain that user and data
centre in same region has low response time while user and
data centre in different region has high response time and the
data centre in hybrid data centre has average response time.
B. Wickremasinghe et.al [4] explain how cloud analyst
works, added features to cloud Sim and show that how cloud
analyst can be used to model real world problem through a case
study of social networking application facebook deployed on
cloud.
M. Malhotra et.al [5] in Simulation for enhancing the
Response and Processing time of Data Centre identified that
deploying two data centre instead of one reduces the response
and processing time. So it is better to deploy two data centre
instead of one data centre.
S. Mohapatra et.al [6] in Comparision of Four Popular
Heuristic for Load Balancing of Virtual Machines in Cloud
Computing explain the static and dynamic algorithm of load
balancing policies. Compare the static policies Round Robin
processing, Throttled scheduling process, Equally spread
current execution load and First come first serve scheduling
process with metrices like response time , processing time and
total cost to fulfil the request. Finally conclude Round Robin
algorithm best for performance.
Ali Naser et.al [7] An Efficient Load Balancing Algorithm
for Virtualized Cloud Data Centre analyse the performance of
Throttled load algorithm with closest data centre, with
optimized response time and with reconfiguring dynamically
w.r.t cost, response time, processing time and showed that
Throttled service broker is best chosen algorithm with better
results in response time and processing time.
Satyanarayanan et.al [8] The Case for VM Based
Cloudlets in Mobile Computing explain the VM (virtual
machine) architecture , in which a mobile user customized
service software on a nearby cloudlet and then uses that
services over a wireless LAN, mobile devices functions as a
thin client relative to server. This strategy is known as cloudlet-
based, resource- rich mobile computing.
IV. PROPOSED MODEL
Researchers have seemed that mobile hardware is resource
poor relative to static client and server hardware. Memory size,
battery life and speed are higher priorities to enhance the
computation. Resource poverty is a major obstacle for
application requiring processing and energy. Satyanarayan et.al
[8] proposed a solution to mobile devices resource poverty is to
use cloud computing. A mobile device can execute a resource
intensive application on a distant high performance compute
server and support thin client user interaction with the
application over the internet. Proposed model state that weak
client can transfer resource intensive application to cloudlet
based on the distance of client to cloudlet. If a private cloudlet
(data centre within the region of user) is present then the
application will transfer to the cloudlet through Wi-Fi,
otherwise application will be executed by public cloud which is
connected to weak client through 3G.
A. Advantage of using Wi-Fi connection
Energy saving
Fast computation i.e. reduce response time
Better user experience.
Using the cloudlet also simplifies the challenge of
meeting the peak bandwidth demand of multiple users.
B. If weak client and cloud are connected through 3G
As distance between client and cloud increases more
energy is consumed because data travel through 3G
For some times CPU sit idle.

Figure 1. Architecture of Cloudlet based Computing
V. EXPERIMENT SETUP
Facebook is one of the social sites getting benefits of cloud
computing with 200million registered users worldwide.
According to survey of 18/06/2009 distribution of the facebook
user base across the globe [4]:

INTERNATIONAL JOURNAL OF TECHNOLOGICAL EXPLORATION AND LEARNING (IJTEL)
www.ijtel.org


IJTEL, ISSN: 2319-2135, VOL.3, NO.3, JUNE 2014 481

TABLE I. DISTRIBUTION OF THE FACEBOOK USER BASE ACROSS THE
GLOBE
Region
Cloud Analyst
Region Id
Users
North America 0
80 million
South America 1
20 million
Europe 2
60 million
Asia 3
27 million
Africa 4
5 million
Ocenia 5
8 million
Here we have analysed the 1/100 scale of facebook user.
TABLE II. 6 USER BASES PARAMETERS REPRESENTING 6 ABOVE MENTIONED REGIONS.
User
Base
Region Time Zone
Peak Hours
(Local Time)
Peak Hours
(GMT)
Simultaneous
Online users
During Peak Hrs
Simultaneous
Online users During Off-Peak Hrs
UB1 0 GMT-6.00 7.00-9.00 pm 13:00-15:00 40,000
4,000
UB2 1 GMT-4.00 7.00-9.00 pm 15:00-17:00 10,000
1,000
UB3 2 GMT+1.00 7.00-9.00 pm 20:00-22:00 30,000
3,000
UB4 3 GMT+6.00 7.00-9.00 pm 01:00-3:00 15,000
1,500
UB5 4 GMT+2.00 7.00-9.00 pm 21:00-23:00 5,000
500
UB6 5 GMT+10.00 7.00-9.00 pm 9:00-11:00 8,000
800
TABLE III. COST CONFIGURATION
Cost per VM per hour(1024 Mb,100MIPS)
$0.10
Cost per Gb of data transfer(from/to Internet)
$0.10
TABLE IV. DATA CENTRE CONFIGURATION
VM image size =1000 VM memory= 1024Mb
VM bandwidth =1000
Data centre architecture =X86

Data centre operating
system= Linux
Data centre VMM = Xen
Number of machine= 20
memory per machine =2048Mb

Storage per machine=
100000Mb
available bandwidth per machine
=10000,

Number of processor per
machine= 4
processor speed =1000 MIPS

VM policy =Time shared
user grouping factor =1000

Request grouping factor
=100
executable instruction per length = 250

TABLE V. LATENCY MATRIX VALUES (MILLISECOND)
Region/Region 0 1 2 3 4 5
0 25.0 100.0 150.0 250.0 250.0 100.0
1 100.0 25.0 250.0 500.0 350.0 200.0
2 150.0 250.0 25.0 150.0 150.0 200.0
3 250.0 500.0 150.0 25.0 500.0 500.0
4 250.0 350.0 150.0 500.0 25.0 500.0
5 100.0 200.0 200.0 500.0 500.0 25.0



TABLE VI. BANDWIDTH MATRIX VALUES (MBPS)
Region/
Region 0 1 2 3 4 5
0 2000.0 1000.0 1000.0 1000.0 1000.0
1000.0
1 1000.0 800.0 1000.0 1000.0 1000.0
1000.0
2 1000.0 1000.0 2500.0 1000.0 1000.0
1000.0
3 1000.0 1000.0 1000.0 1500.0 1000.0
1000.0
4 1000.0 1000.0 1000.0 1000.0 500.0
1000.0
5 1000.0 1000.0 1000.0 1000.0 1000.0
2000.0

Scenario 1: In this scenario we have considered the 6
user bases with single data centre (present only one region at
a time) in region 0, region 1, region 2, region 3, region 4 and
region 5. The table given below show the response time of
user bases according to the position of data centre in region
0, region 1, region 2, region 3, region 4, region 5
respectively.









INTERNATIONAL JOURNAL OF TECHNOLOGICAL EXPLORATION AND LEARNING (IJTEL)
www.ijtel.org


IJTEL, ISSN: 2319-2135, VOL.3, NO.3, JUNE 2014 482

TABLE VII. RESPONSE TIME OF 6 USER BASE AFTER SIMULATION BY PLACING DATA CENTRE IN EACH REGION RESPECTIVELY.
DATA
CENTRE REGION
RESPONSE TIME
of UB0
RESPONSE
TIME of UB1
RESPONSE
TIME of UB2
RESPONSE
TIME of UB3
RESPONSE
TIME of UB4
RESPONSE TIME of
UB5
0 3087.52 1699 2626.4 2106.7 4820.65 5719.24
1 3824.81 153.37 2856.6 2599.2 5042.98 5964.65
2 3335.53 1982.83 2347.24 1876.82 4640.6 5959.57
3 3503.7 2449.51 2628.35 1684.91 5318.93 6503.96
4 3519.19 2121.74 2608.07 2605.14 4387.59 6521.04
5 3220.04 1852.62 2777.28 2637.47 5304.43 5681.6


Figure 2. User base response time (milliseconds) according to position of
single data centre in a region
We can analyse from the graph that
UB0 has least response time when data centre is in
region 0 as compared to when data centre is in
region 1, 2, 3, 4, 5 respectively.
UB1 has least response time when data centre is in
region 1 as compared to when data centre is in
region 0, 2, 3, 4, 5 respectively
UB2 has least response time when data centre is in
region 2 as compared to when data centre is in
region 0, 1, 3, 4, 5 respectively.
UB3 has least response time when data centre is in
region 3 as compared to when data centre is in
region 0, 1, 2, 4, 5 respectively.
UB4 has least response time when data centre is in
region 4 as compared to when data centre is in
region 0, 1, 2, 3, 5 respectively.
UB5 has least response time when data centre is in
region 5 as compared to when data centre is in
region 0, 1, 2, 3, 4 respectively.
Scenario 2: User base and data centre configuration is
same but here we have varied the number of data centre from
one to four to observe the performance.
Case 2.1: Single data centre: In this approach
application is deployed at a single location, here it is at
region 0 i.e at North America. After simulation table given
below shows the average, min, max response and
processing time
TABLE VIII. OVERALL RESPONSE TIME SUMMARY
Avg
(ms)
Min (ms) Max (ms)
Overall
Response time:
4252.81 263.09 15170.60
Data Centre
Processing time:
3861.29 13.14 14715.05
TABLE IX. DATA CENTRE REQUEST SERVICING TIME
Data Centre Avg (ms) Min (ms) Max (ms)
DC1 3861.29 13.14 14715.05

Case 2.2: Two data centre: In this approach application is
deployed at two locations but within the same region, here
these are at region 0 i.e at North America. After simulation
table given below shows the average, min, max response
and processing time.
TABLE X. OVERALL RESPONSE TIME SUMMARY
Avg (ms) Min (ms) Max (ms)
Overall response time: 2313.78 232.13 8160.86
Data Centre processing time: 1979.34 13.14 7678.95
TABLE XI. DATA CENTRE REQUEST SERVICING TIME
Data Centre Avg (ms) Min (ms) Max (ms)
DC1 1920.30 13.15 7678.95
DC2 2032.69 13.14 6787.86
0
1000
2000
3000
4000
5000
6000
7000
UB0 UB1 UB2 UB3 UB4 UB5
region 0
1
2
3
4
5
INTERNATIONAL JOURNAL OF TECHNOLOGICAL EXPLORATION AND LEARNING (IJTEL)
www.ijtel.org


IJTEL, ISSN: 2319-2135, VOL.3, NO.3, JUNE 2014 483

Case 2.3: Three data centre: In this approach application
is deployed at three locations but within the same region,
here these are at region 0 i.e at North America. After
simulation table given below shows the average, min, max
response and processing time.
TABLE XII. OVERALL RESPONSE TIME SUMMARY
Avg (ms) Min (ms) Max (ms)
Overall response time: 1737.31 107.97 6013.18
Data Centre
processing time:
1420.73 13.14 5500.36
TABLE XIII. DATA CENTRE REQUEST SERVICING TIME
Data Centre Avg (ms) Min (ms) Max (ms)
DC1 1450.29 13.15 5500.36
DC2 1464.16 13.14 5122.78
DC3 1348.65 25.09 4714.27

Case 2.4: Four data centre: In this approach application is
deployed at four locations but within the same region, here
these are at region 0 i.e at North America. After simulation
table given below shows the average, min, max response
and processing time.
TABLE XIV. OVERALL RESPONSE TIME SUMMARY
Avg (ms) Min (ms) Max (ms)
Overall response time: 1524.65 142.91 6004.21
Data Centre processing time: 1214.06 13.14 5504.65
TABLE XV. DATA CENTRE REQUEST SERVICING TIME
Data Centre Avg (ms) Min (ms) Max (ms)
DC1 1220.34 13.14 4286.23
DC2 1177.69 13.15 4669.59
DC3 1148.22 46.98 3269.44
DC4 1313.90 25.09 5504.65
TABLE XVI. RESPONSE TIME AND PROCESSING TIME OF DATA CENTRE
WHEN THE NUMBER OF DATA CENTRE INCREASES.
Number of Data
centre
Response time
(millisecond)
processing time
(millisecond)
DC1 4252.81 3861.29
DC2 2313.78 1979.34
DC3 1737.31 1420
DC4 1524.64 1214.06


Figure 3. Response time, Processing time VS number of data centre
Graph shows that increasing the number of data centre
decreases the response time and processing time.
VI. OBSERVATION
Figure 3 shows that bringing services closer to user
bases reduce the response time (improves the quality
of service).
Figure 4 shows that we can improve quality of
service (response time and processing time) by
increasing the number of data centre.
Energy saving as response and processing time
reduces.
Fast computation i.e .reduces response time.
Better user experience.
Using the cloudlet also simplifies the challenge of
meeting the peak bandwidth demand of multiple
users.
VII. CONCLUSION
Companies all over the world are financing or utilizing
the services of cloud computing anywhere and at anytime.
Cloud computing consist of thousands of data centre.
Resources and services change with user requirements and
environments. Data centre services, users and process are
significant for availability and better quality of services. This
research concludes that hierarchical representation of data
centre leads to better user experience and save energy by
bringing services closer to user bases that reduce the
response time (improves the quality of service) and also
analyze that we can improve quality of service (response
time and processing time) by increasing the number of data
centre.
REFERENCES
[1] Dr. Neeraj Bhargava, Dr. Neetu Bhargava International journal of
Computer Science and Mobile Computing, Performance Analysis of
Cloud Computing for Distributed Client. IJCMC, vol. 2, issue 6,
june2013, ISSN 2320-088X.
[2] Ashraf Zia, M.N.A. khan IJ Modern Education and Computer
Science, A Scheme to reduce response time in cloud computing
environment. Published online july2013 in MECS.DOI:
10.5815/ijmecs.2013.06.08.
[3] Dhaval Limbani, Bhavesh Oza/ International journal of Engineering
Research and Applications IJERA, Proposed Service Broker
Strategy in Cloud Analyst for Cost Effective Data Centre Selection.
[4] Bhathiya Wickremasinghe, Rodrigo N. Calheiros, and Rajkumar
Buyya, Cloud Analyst: A cloudSim-based Visul Modeller for
Analysing Cloud Computing Environments and Application.
[5] Manish Malhotra Intrenational journal of computing and corporate
research, A Simulation for Enhancing the Response and Processing
Time of Data Centre volume 1issue 3 manuscript 8 November 2011.
[6] Subasish Mohapatra, K Sumruti Rekha, Subhadarshini Mohanty,
Comparision of Four Popular Heuristic for Load Balancing of
Virtual Machines in Cloud Computing International journal of
computer applications(0975-8887) Volume 68-No. 6, April 2013.
[7] Ali Naser Abdulhussein, Jugal Harshvadan Joshi, An Efficient Load
Balancing Algorithm for Virtualized Cloud Data Centre.
[8] Mahadev Satayanaranan, Paramvir Bhal, The Case for VM Based
Cloudlets in Mobile Computing published by IEEE CS 1536-
1268/09/2009 IEEE.

0
1000
2000
3000
4000
5000
DC1 DC2 DC3 DC4
response
time(milliseco
nd)
processing
time
(millisecond)