AN ENHANCE LOAD BALANCING ALGORITHM FOR HETEROGENEOUS DATA CENTERS Authors: S.E.C. Nanayakkara, N.T. Mapa, D.S.A. Kandawala, H.P.P.A. Pemarathna November 2014
AN ENHANCE LOAD BALANCING ALGORITHM FOR
HETEROGENEOUS DATA CENTERS
Authors: S.E.C. Nanayakkara, N.T. Mapa, D.S.A. Kandawala, H.P.P.A. Pemarathna
November2014
ii
Abstract
Load balancing is the process of distributing the load
among various nodes of a distributed system. The distribution
mechanism relies on underlying load balancing algorithm in
use. This concept called load balancing arises due to the fact
of requiring a proper resource utilization mechanism to gain
the maximum result and performance out of the resources
available. Many research projects have been conducted in this
field, but still it is in need of further updates because of
the rising user base and increasing need of services.
This project mainly focuses on the dynamic environmental
behaviors exist in the heterogeneous server farms to implement
a better performing novel load balancing algorithm. In advance
the algorithm follows a simple and concise flow of procedure
to generate the best possible decisions.
This novel algorithm is motivated by the significant
characteristics identified in both the Round Robin algorithm
(RR) and Weighted Round Robin algorithm (WRR). A comparative
analysis between other extended version of Round Robin
algorithms and Modified RR is included in this report as to
clearly clarify the significant advantages. A detailed
analysis of associated qualitative parameters namely Overall
Response time, Data Center Processing time and Grand total
cost was taken into consideration as performance metrics. All
the associated test results were generated and analyzed using
the java based simulator called Cloud Analyst which provides
necessary features and tools. Furthermore this report contains
the directions for future works that can be done related to
this project.
ii
The ultimate goal of this project is to implement an
innovative and enhanced load balancing algorithm for
heterogeneous data centers while show casing it stands ahead
of other algorithms with clear statistical and figurative
evidence by the use of test results obtained after application
of the algorithm on a simulated environment.
3
Ackn ow ledgemen ts
First and foremost, we would like to thank the
supervisor in charge of this project, Mr. Lahiru
Ariyasinghe and the Head of Department Mr. Anuradha
Jayakody for the valuable guidance and advices. They
inspired us greatly to work in this project. Their
willingness to motivate us contributed tremendously to
our project. We also would like to thank them for showing
us some example that related to the topic of our project.
Besides, we would like to thank the Department of
Computer Systems And Networking for providing us with a
good environment and facilities to complete this project.
Also, we would like to take this opportunity to thank to
both universities for offering this subject, Computer
Technology Project.
Finally, an honorable mention goes to our families and
friends for their understandings and supports on us in
completing this project. Without helps of the particular
that mentioned above, we would face many difficulties
while doing this project.
4
Table Of Contents Declaration................................................................................................................................i Abstract....................................................................................................................................ii Acknowledgements.................................................................................................................iii List Of Figures.........................................................................................................................v List Of Tables........................................................................................................................ vii1. INTRODUCTION.......................................................................................................... 11.1 Problem to be addressed........................................................................................... 11.2 Background Context................................................................................................. 11.3 Research Gap........................................................................................................... 51.4 Research Question.................................................................................................... 6
2. LITERATURE REVIEW.............................................................................................. 72.1 Round Robin with Server Affinity[1]...................................................................... 7
5
2.2 Fair Round Robin[2]................................................................................................ 82.3 Efficient & Enhanced Algorithm............................................................................. 9
3. METHODOLOGY....................................................................................................... 103.1 Calculating Weights for Virtual Machines............................................................. 113.2 Prioritizing the Virtual Machines........................................................................... 123.3 Checking the Availability........................................................................................ 13
4. RESEARCH FINDINGS............................................................................................. 145. RESULTS & DISCUSSION........................................................................................ 155.1 Cloud Analyst Simulator[4]................................................................................... 155.2 Performance Metrics.............................................................................................. 165.3 Testing Procedure.................................................................................................. 175.4 Test Results............................................................................................................ 185.5 Discussion on Generated Results........................................................................... 24
6
6. CONCLUSION AND FUTURE WORKS................................................................. 25REFERENCE ....................................................................................................................... 27
7
List Of Figures
Figure 2:1:1 Algorithm Comparison ...................................................................7
Figure 2:2:1 Round Robin Algorithm[2]...............................................................8
Figure 2:2:2 Fair Round Robin Algorithm[2]........................................................8
Figure3:1 Four main functions of the Modified Round Robin algorithm...........................................................................................................10
Figure3:1:1 The Equation for calculating the weight.........................................11
Figure3:2:1 Virtual Machine List........................................................................12
Figure3:2:2 weight1[] array...............................................................................12
Figure3:2:3 weight2[] array...............................................................................12
Figure3:2:4 Orderarr[] array..............................................................................12
Figure3:2:5 Java code for prioritizing the VirtualMachines...........................................................................................................13
Figure3:3:1 Pseudo code for Checking the
8
Availability.........................................................................................................13
Figure4:1 Flow chart for the Functionality of the Modified Round Robin LoadBalancing Algorithm..........................................................................................14
Figure5:1:1 Cloud Analyst Main Window..........................................................15
Figure5:1:2 Cloud Analyst Main ConfigurationWindow............................................................................................................16
Figure5:3:1 Summary of the Testing Procedure...............................................17
Figure5:4:1 User Base Parameter List across the regionsof theGlobe................................................................................................................18
Figure5:4:2 The Data Center and VM Settings used forthe simulation........................................................................................................18
Figure5:4:3 Main Configuration for Scenario1 ...............................................19
Figure5:4:4 Main Configuration for Scenario 2................................................19
Figure5:4:5 Result of Overall Response Time Compared with ServerAffinity.............................................................................................................20
9
Figure5:4:6 Result of Data Center Processing Time Compared with ServerAffinity......................................................................................................................20
Figure5:4:7 Main Configuration for Comparison with Fair RoundRobin........................................................................................................................21
Figure5:4:8 Result of Overall Response Time Compared withFair RoundRobin........................................................................................................................21
Figure5:4:9 Result of Data Center Processing Time Compared with Fair Round Robin.................................................................................................................................22
Figure5:4:10 User Base wise Overall Response Time Compared toRR.............................................................................................................................22
Figure5:4:11 User Base wise Data Center Processing Time Compared toRR............................................................................................................................23
Figure5:4:12 User Base wise Grand Total Cost Compared toRR............................................................................................................................23
vi
List Of Tables
Table 1:2:1 Literature Survey on existing Load Balancing Algorithms..........................3
Table 1:2:2 Literature Survey on existing Load Balancing Algorithms 2nd stage..........5
1
1.INTRODUCTION
1.1 Problem to be addressed
With the consistently growing popularity of cloud and
number of cloud service providers, it is obviously most
significant to upgrade the cloud services which are
provided by cloud service providers by improving the
performance. Hence to satisfy the requirements regarding
bandwidth, memory requirements and response time of the
user base request can be done by implementing efficient VM
Load Balancing policies.
Currently there are only a limited number of VM load
balancing policies available and also follows Sophisticated
and complicated techniques to balance the load among
virtual machines in a data center. Hence this project
introduces an appropriate solution which addresses on
enhancing a proper load balancing algorithm by considering
in hand of simplicity, reliability and practical
aspects.This problem statement leads to the broad
objectives that are summarized below.
Evaluate and analyze the different existing loadbalancing algorithms and enhance a selected algorithmto implement a novel version which is capable ofhandling for heterogeneous data centers.
To implement the algorithm on Cloud Analyst and to analyze its performance
1
Comparison of the proposed algorithm with the existingalgorithms on identified parameters
The rest of the sections more concern to satisfy theabove mentioned objectives.
1.2 Background Context
This section mainly includes the literature survey
analysis and related work done related in the field of load
balancing. In this stage the main focus was to compare and
contrast the different load balancing algorithms that has
been evolved throughout past years and to conduct in-depth
and clear view analysis of those to choose a better path
regarding this proposed project. The in detailed
information about the conducted analysis is summarized as
follows.
Algorithm Paper
Drawbacks Advantages
ESCE Algorithm
Analysisof loadbalancingalgorithm incloudcomputing &study of game
Fault tolerance low Throughput,response time,overhead in alower levelcompared to the
otherdynamicEfficient
loadbalancingalgorithm incloud environment
The jobs areequally spread,thecomplete,computingsystem isloadbalancedand numberof VMs areunderutilize. Hence
A survey onloadbalancing algorithms in
Single point of failure
Efficient andenhancedalgorithm in
Additional computation overhead to scanqueueagain and again
Weighted least connectionAlgorithm
Anefficientloadbalancingscheme forheterogeneousprocessingsystem
Evaluate node load balancing by measuringof connect quantity, the method is not perfect. There are otherparameters that affect
May allocate the taskforeachnodeaccordingtoconnectionquantity
Show the An adaptive
weightleast
The load status of servers is not
2
algorithm based onserver clusters
status can’t be revealedby the number of links only.
The weightdescription is
inaccurate.In traditionalload balancingalgorithm,server weight issupposedas a
Ant colony optimization Algorithm
Antcolonyoptimizationfor effectiveloadbalancing incloudcomputing
Networkoverhead becauseof the largenumber of ants.
Onlyavailabilityof node is beenconsidered.
Nodes statuschange after
Optimum solution ofload.
No singlepoint of failure.
Honey bee inspired load balancing Algorithm
A survey onloadbalancing incloud
Low priority load become staycontinuously
Maximizing the throughput
Low overheadLoad
balancing incloud
Throughput is not increased with an
ThrottledAlgorithm
Comparativeanalysis of existing dynamic load
When requests increase beyond threshold valuewaiting timewill beincreased as the
Better response time.
Round RobinAlgorithm
Loadbalancing ofweb serversystem using
Does not consider about the current load andresponsiveness Efficient
loadbalancingalgorithm in
Fast and easy to implement.
Table 1:2:1 Advantages and disadvantages of some of the existing load balancing algorithms
3
4
After performing the above analysis the initial aim was to
focus on developing an enhanced load balancing algorithm
while mitigating the errors and drawbacks identified during
the analysis to achieve better goals in server farms on any
open environment. Along with the literature survey few most
important and compulsory qualitative factors were
identified which should be considered towards better load
balancing. Further in depth details regarding the above
mentioned factors will be described in the sections to
come.
From the above literature survey according to the
gathered information this project proposes to enhance the
Round Robin (RR) load balancing algorithm. According to the
analyzed survey information it clearly highlights the Round
Robin algorithm as the efficient algorithm suited for a
light weight server environment compared to the otheralgorithms due to itssimplicity and reliability.
WLC Throttled Honey
BeeACO ESCE
CompareWith
RR
6
The second section directs a comparative analysis regarding
the work done related to this research area by the other
parties to enhance the Round Robin load balancing algorithmwith the same objective.
Algorithm Name
Description Plus Points DrawbacksRound Robinwith ServerAffinity [1]
Implemented usingHash maps and VMstate list.
Save the stateOf the previous allocation of a VM to a
Only consideredVMs with homogeneous
Fair RoundRobin (FRR) [2]
Provide fairness toexecuting nodes by considering
Considers thevariations of request size.
Only consideredVMs with homogeneous Efficie
nt &EnhancedAlgorit
Maintains VM indextable containing state of
Checking VMavailability& consideringexpected Response
Problem ofdeadlocks & Server overflow.
1.3 ResearchGap
Table 1:2:2 Survey on existing algorithms 2nd stage
According to the previous section there were drawbacks
and problems remaining with the existing and newly
researched and proposed load balancing algorithms. They can
be listed as follow
They assume that the virtual machines in a datacenter have homogeneous characteristics.
Deadlocks can be occurred due to actions of some loadbalancing algorithms.
7
Some of the algorithms does not consider about the availability of the virtual machines in a datacenter.
The Modified Round Robin bridge the gap of these
drawbacks remaining with the existing load balancing
algorithms.
8
1.4 Research Question
Due to rapid changes of networking environments, such
as increasing the user base, number of requests generated
and expected services from the servers. The load balancing
algorithms has to evolve and need to adapt the on-going
transformation. Hence implementing an applicable algorithm
still prevails in the research state.
Further questions may arise when existing algorithms
follow complex and mature techniques to balance the load
among networking data centers. There are various issues
while dealing with load balancing in a cloud computing
environment. Each load balancing algorithm must be such as
to achieve the desired goal. Some algorithms aims at
achieving higher throughput, some aims at achieving minimum
response time, some other aims to achieve maximum resource
utilization while some aims at achieving a trade-off
between all these metrics. Hence come up with an efficient
and accurate Load balancing still being required in this
field of research area.
9
2. LITERATUREREVIEW
2.1 Round Robin with Server Affinity[1]
Basically this algorithm addresses the problem, which is
not saving the state of previous allocation of VMs to a
request from a given User base. If any request received
from same user base the algorithm must run again. That
clearly shows the increment of the response time and the
processing time of the request. As a convenient solution RR
with Server Affinity aims to reduce total response time and
processing time of user base requests by Enhancing Round
Robin Algorithm.
Implementation of the RR Sever Affinity is accomplished using hash map and a VM
statelist.
Hash Map: Used to store the entries for the last VM allocated to the request. VM State List: Stores the allocation status of each VM.
When a request comes from user base, if an entry for the
given user base exist in the hash map and if particular VM
is available in the VM state list, no need to run algorithm
again. It leads to reduce the response time and processing
time significantly. That has been confirmed by the
10
following graph. The main fact about the algorithm is which
implemented only for homogenous networking environment[5].
Figure 2:1:1 AlgorithmComparison
11
2.2 Fair Round Robin[2]
This algorithm identifies the existing algorithms as
complex and increase executing node's load which degrades
its performance. As a solution this algorithm enhances the
Round Robin algorithm by implementing dynamic time quantum
based on algorithm executing round. Hence this algorithm
follows quit simple strategies.
The procedure of this algorithm mainly concerned with fair
scheduling when the burst time of incoming request load is
having great variance. The overall algorithm execution
requires less number of rounds to follow thus the algorithm
when applied to large number of request performs faster
providing better response time to any request. Fair Round
Robin algorithm thus provides fairness to larger requests
and smaller one also to the load coming at executing node
in the cloud. Same as the Server Affinity this also
implemented for VMs with homogenous characteristics[2].
Following results shows the better performance of the FairRR algorithm than RoundRobin algorithm.[6]
Figure 2:2:1 Round Robin Algorithm[2]
13
2.3 Efficient & Enhanced Algorithm
This algorithm also proposes an enhanced Round Robin
Algorithm by comparing the performance with existing
algorithm characteristics. Hence this directs a comparative
analysis to address as an efficient algorithm[3].
This algorithm follows the procedure, which is to find the
expected response time. To calculate the response time
following equation has been used.
Response Time = Fint - Arrt + TDelay ………..(1)
Where, Arrt is the arrival time of user request and Fint is
the finish time of user request and the transmission delay
can be determined using the following formulas;
TDelay = Tlatency + Ttransfer……. (2)
Where, TDelay is the transmission delay T latency is the
network latency and T transfer is the time taken to
transfer the size of data of a single request (D) from
source location to destination.[7]
Ttransfer = D / Bwperuser ……….. (3) Bwperuser = Bwtotal / Nr ……….. (4)
Where, Bwtotal is the total available bandwidth and Nr is
the number of user requests currently in transmission. The
Internet Characteristics also keeps track of the number of
14
user requests in between two regions for the value of
Nr[3].
As a disadvantage this algorithm can be fall in to failures
like starvation and server overflow. Other than that VMs
contain equal characteristics which are known as homogenous
datacenters.[8]
15
3. METHODOLOGY
Based on the literature survey that has been conducted,
this project came up with the modified algorithm which
addresses the most of the drawbacks remaining with the
existing load balancing algorithms. The proposed modified
load balancing algorithm is inspired by the Round Robin
algorithm and is concerning more towards the heterogeneous
characteristics of the Virtual Machines in a Data Center
such as memory, bandwidth and also the availability of the
Virtual Machines. The Modified Round Robin algorithm canbe mainly split into four functions as given below.
CalculatingWeights usingBandwidth and
Memory
Prioritize VMs
Check the Availability of the VMs
16
Select the VMsin a
Round Robinmanner
Figure3:1 Four main functions of the Modified Round Robin algorithm
17
First the all the algorithm calculates weights for allthe Virtual Machines in a Data
Center based on the VMs Bandwidth and Memory. And then it prioritizes the Virtual Machines according to
their weights. Higher weights gain higher priority. Then the algorithm checks for the availability of the Virtual Machines. And finally it will select the Virtual Machine based on the priority and the availability
3.1 Calculating Weights for Virtual MachinesAs mentioned earlier, weights are calculated for
all the Virtual Machines in a Data Center. Theweight was calculated by dividing the addition ofthe Bandwidth and Memory of the Virtual Machine bytwo into hundred.
B M W
2 *100Figure3:1:1 The Equation for calculating the weight
B = Bandwidth of the VM M = Memory of the VMW = Weight of the VM
18
After going through many number of tests andsimulations, according to their test results, itshowed that Bandwidth and Memory has the same effecton the performance of a Virtual Machine. Hence theaverage value of the bandwidth and memory has beentaken. As Bandwidth and Memory variables store longvalues, to simplify the weight value further theaverage value of the addition of bandwidth andmemory has been divided by 100.
One of the main important point is that all thecalculate weights are stored in an array calledweight1[9].
19
3.2 Prioritizing the Virtual Machines
The functionality of prioritizing the Virtual Machines can be presented in diagrams as shown below.
First the Virtual Machine List will be prepared according to the GUI inputs
VM List
Figure3:2:1 Virtual Machine List
Then according to the first function which has briefly explained earlier, all the calculated weights will be stored in the weight1[] array.
Weight1[]
Figure3:2:2weight1[] array
For an example assume that K2 > K0 > K1 > K3 As the next step the algorithm will take a copy
of the weight1[] array(weight2[]) and then weight2[] array will be
sorted in descending order.Weight2[]
Figure3:2:3 weight2[] array
And then by storing weight1[]s index of weight2[]s element, another array called orderarr[] will be prepared.
Orderarr[]
21
Finally the Orderarr[] array will contain all the indices of Virtual Machine
List in descending order based on their weights. The corresponding java code for this function can be shown as below.
for(int z=0; z < weight2.length; z++){
orderarr[z] = Arrays.asList(weight1).indexOf(weight2[z]);
}Figure3:2:5 Java code for prioritizing the Virtual Machines
3.3 Checking the Availability
This function mainly checks the availability of
the Orderarr[] arrays elements in a Round Robin
manner. For further explanation a pseudo code canbe presented as below.
Checking the availability of orderarr[currVM]th VM; Initially currVM = -1Loop{
currVM++ Availabilityof the
orderarr[currVM]
22
VMif(availa
ble)select the
VM;else continue;
}
Figure3:3:1 Pseudo code for Checking the Availability
23
The CurrVM variable is a global variable which
represents the indices of the Orderarr[] array.
4. RESEARCH FINDINGSSt
Calculate weightsfor alltheVMs
PrioritizeVMs
currVM++
YesIs
orderarr[currV M]th VM
Return
orderarr[currVM]
No
currVM++ St
currVM =0
YesIs
currVM
No
Figure4:1 Flow chart for the Functionality of the Modified Round Robin Load Balancing Algorithm
24
5. RESULTS & DISCUSSION
5.1 Cloud AnalystSimulator[4]
After implementing the code and performing the necessary
modifications, the modified version of Round Robin load
balancing algorithm which is described in this report was
tested using a simulated environment. For the simulation
purpose Cloud Analyst simulator had been used which is
java based application tool including all the important
functionalities dedicated for testing load balancing
algorithms.[10]
It provides a simple and generic graphical interface
with all the options and settings needed to build a
simulated environment anyone desires. [11]
26
Figure5:1:2 Cloud Analyst MainConfiguration Window
5.2 Performance Metrics
Before conducting the testing phase three main performance
parameters were selected. The test results was analyzed
against the below listed performance metrics.
• Overall Response Time :Time between the submission of the request and the receipt if response.
• Data Center Processing Time :Time consumed to process the entire user request
• Cost Factor :This is calculated as the addition of both VM Cost
and Data Transfer Cost. Mainly the VM cost is
related with the memory consumption, storage
consumption and VM up time.
28
5.3 Testing Procedure
The overall testing procedure can besummarized as shown below.
Figure5:3:1 Summary of the Testing Procedure
Mainly the testing procedure can be divided into 3
categories as mentioned below. (1) Comparison of
Server Affinity with Modified Code(2) Comparison of Fair Round Robin with Modified Code(3) Comparison of Generic Round Robin Code with Modified Code.
The Efficient and Enhanced Algorithm was taken as reference
to gather knowledge on how to perform a well-defined
testing flow and to learn on theories behind evaluating
load balancing algorithms. It was taken into account for
the comparative analysis as it is not a Pio Round Robin
based load balancing algorithm. But the other two
algorithms namely Fair Round Robin and Server Affinity is
30
5.4 Test Results
(1) Comparison of with Modified Code
For this comparison two different scenarios was taken
into consideration. These scenarios described as
follows.
Scenario1 : Six User Bases with only one Data Center Scenario2 : Six User Bases with two Data Centers
Figure5:4:1 User Base Parameter List across the regions of the Globe
33
Figure5:4:5 Result of Overall Response Time Comparedwith Server Affinity
Figure5:4:6 Result of Data Center Processing TimeCompared with Server Affinity
34
(2) Comparison of Fair Round Robin with Modified Code
All the necessary VM setting and test resultsare clearly include below.
Figure5:4:7 Main Configuration for Comparison with Fair Round Robin
Figure5:4:8 Result of Overall Response Time Comparedwith Fair Round Robin
35
Figure5:4:9 Result of Data Center Processing Time Compared with FairRound Robin
(3) Comparison of Generic Round Robin with Modified Code
For this comparison the same VM setting and main
configuration settings which was used with the
comparison between Fair Round Robin was used. But to
make it more efficient and to generate more practical
related results the number of user bases was increased
gradually as to increase the amount of requests
directed towards the data center. This modification
was done to test the behaviour of the modified code
against extreme conditions.
37
Figure5:4:11 User Base wise Data Center Processing Time Compared to RR
Figure5:4:12 User Base wise Grand Total Cost Compared to RR
38
5.5 Discussion on Generated Results
A clear look into both the comparisons performed with the
Server Affinity and Fair Round Robin load balancing
algorithms shows that the modified algorithm which is
thoroughly described throughout this report shows best
results with respect to Overall Response Time and Data
Center Processing Time.
The main reason for this outcome is that the modified code
always considers the ability or the power of the server
when directing the traffic towards them. But both the
Server Affinity and Fair Round Robin do not take the server
capabilities into consideration. And the other reason can
be stated as the simplicity of the algorithm which reduced
the computational power and leads to much lesser Data
Center Processing values. This concludes that the modified
algorithm performs better compared to the other extended
versions of the basic Round Robin load balancing algorithm.
Furthermore to make it clearer and to improve clarity of
the testing phase the final comparison performed against
the basic Round Robin shows favourable results towards the
modified Code. The reason that should be highlighted is
that using the Modified Code, the server capabilities and
server availability is taken into consideration. And it is
modified in a simpler way giving a twist to the generic
39
Round Robin code which leads to the outcome of better
results.
As the final statement we can conclude that the Modified
Code achieved all is expected aspects and stand ahead of
the algorithms which are taken into consideration in this
report. Moreover the implementation flow of the Modified
code has opened different directions of enhancing the load
balancing algorithms to make it perform better.
40
6. CONCLUSION AND FUTURE WORKS
Regardless of developments in the IT industry, Cloud
computing is an expanding computing paradigm which has
influenced in every field of entity in the globalize
industry whether it is in the public or private sector. Due
to the fast growing cloud importance, discovering new
methods to improve cloud service is an area of concern and
research focal point.
In current situation, there are only a limited number
of VM load balancing policies available. Besides that Cloud
Analyst implement the characteristics of VMs inside a data
center in equal manner which is an impractical solution for
a real world scenario. The research work engaged
developing and enhancing a Load Balancing algorithm for the
cloud environment and directing a comparative analysis of
proposed algorithm with existing proposed load balancing
algorithms. Along with the observations most of the
researches enhanced the performance of load balancing
algorithm for homogeneous VM data centers implement using
Cloud Analyst.
The major contribution of the proposed work in the field of
cloud computing is that the authors have introduced a
upgraded algorithm functionality, that is to change the
characteristics of VMs in a data center and come up with
an equation to change the VM characteristics. Further
41
proposed algorithm provides the optimal and most specific
solution to allocate the work load among VMs to increase
the performance factors that is to say response time and
processing time in an efficient manner.
Without ending this research at this point, it can be
extend by doing further modifications. The proposed
algorithm can be upgraded by adding the energy awareness by
considering the available bandwidth, memory and other
necessary parameters dynamically. As the proposed algorithm
consider about the prioritized servers there can be
situations like server overflow and starvation. By solving
these problems this proposed algorithm can be turned in to
a more efficient and accurate load balancing algorithm.
Moreover this proposed algorithm was tested in a virtual
environment running in a simulator. As a new
42
step the proposed algorithm can be tested in a real
environment. Moreover the failure handling mechanisms such
as handling exceptions can be added in to the simulation
when a server goes down for further enhancements.
43
REFERENCES
[1] Komal Mahajan et al., “Round Robin with ServerAffinity: A VM LoadBalancing Algorithm for Cloud Based Infrastructure”, Dept.CSE & ICT, Jaypee Univ. ofInformation Technology, Waknaghat,India, July 2013.[2] Stuti Dave and Prashant Maheta, “Utilizing Round RobinConcept for Load BalancingAlgorithm at Virtual Machine Level in Cloud Environment”, BH Gardi College of Eng.& Tech., Gujarat,India, May 2014.[3] Tejinder Sharma and Vijay Kumar Banga, “Efficient andEnhanced Algorithm inCloud Computing”,Mar. 2013.[4] Bhathiya Wickremasinghe, “CloudAnalyst: A CloudSim-
based Tool for Modelling and Analysis of Large Scale Cloud
Computing Environments”, Dept. CSSE, Melbourne Univ., June
2009.
[5] (2014, Mar. 6). LoadBalancers [Online]. Available:https://wiki.gogrid.com/index.php/Load_Balancers#Load_Balancer_Algorithm[6] Shilpa S.et al., “Analysis of Load Balancing Algorithms
in Cloud Computing and study of Game Theory.” International
Journal of Advanced Research in Computer Engineering
44
& Technology (IJARCET),vol.3, Apr. 2014[7]AkshayDaryapurkarand Mrs.V.M.Deshmukh. “Efficient Load
Balancing Algorithm in Cloud Environment” International
Journal of Computer Science and Applications, vol.6, Apr.
2013
[8] Brajendra Kumar and Dr. VineetRichhariya. “Load
Balancing of Web Server system using service queue length”
International Journal of Emerging Technology and Advanced
Engineering, vol.4, May 2014
[9] Xiaonian Tong and Wanneng Shu. “An efficient dynamicload balancing scheme forheterogeneous processing system”. ComputationalIntelligence and National computing,2009, 319-322.
45
[10] David Quaid. (2013, Jun 06). Load Balancing Scheduling
Methods explained [Online]. Available:
http://www.loadbalancerblog.com/blog/2013/06/load-
balancing- scheduling-methods-explained
[11] Dr. Hermant S. Mahalla et al., “Load Balancing onCloud Data Centers,” InternationalJournal of Advanced Research in Computer Science andSoftwareEngineering, vol.3, Jan.2014