International Conference on Advances in Computing, Communication Control and Networking (ICACCCN2018)

HEURISTIC LOAD BALANCING ALGORITHMS

IN VULNERABLE CLOUD COMPUTING
ENVIRONMENT

Nidhi Verma Dr. Vidushi Sharma

School of ICT School of ICT (Assistant Professor)
Gautam Buddha University Gautam Buddha University Greater
Greater Noida-201308, India Noida-201308, India
[email protected] m [email protected]

Monika Kashyap Apurva Jha

School of ICT School of ICT
Gautam Buddha University Gautam Buddha University
Greater Noida-201308, India Greater Noida-201308, India
[email protected] m [email protected] m

Abstract— Rapid innovation in IT sector has led to the and service and so forth as an utility. It can enhance its
development of cloud computing. Cloud Computing is a accessibility assets. Cloud Computing has many
general term for giving access to set of administrations advantages, some of them are: cost saving, flexibility,
over the web. Those administrations incorporate mobility, insight, increased collaboration, damage
networks, servers, storage, applications, compute power prevention, damage recovery, sustainability and many
other IT related assets. As cloud computing is sought more. Along with advantages it has some disadvantages
after now-a-days, with the expansion for the cloud like security issues, accessibility, bandwidth, load
services, load on cloud is increasing excessively. So to balancing job scheduling.
lesser the unfriendly impacts of vulnerable computing
condition, we have to adjust the load. In this paper, On the basis of accessibility, cloud computing is
Honey bee behavior inspired load balancing (HBB-LB) is categorized into-
implemented. Further, to lesser the effect of drawbacks 1) Public Cloud- In this type third party delivers shared
of HBB, we have proposed Ant Colony Optimization assets, system, and capacity to clients on the web. For
based Load Balancing Algorithm. Also, we have example- Salesforce.com. Public Cloud Architecture is
performed comparison for the algorithms considering characterized into service models. Those service models
some QoS factors. are as follows:-
x SaaS- SaaS stands for Software as a service. SaaS
functions at the top most level. SaaS adds to cloud
Keywords— Cloud Computing; Load Balancing; HBB- computing by delivering vendor’s cloud-based software to
Inspired Load Balancing Algorithm; Ant Colony end users.
Optimization Load Balancing Algorithm; x PaaS- PaaS stands for Platform as a service. PaaS
functions at the lower level than SaaS. PaaS adds to cloud
computing by providing user a platform for cloud
where user can develop, manage and deliver services.
I.INTRODUCTION x IaaS- IaaS stands for Infrastructure as a service. IaaS
Cloud computing is a standout amongst the most developing functions at the lowest level. IaaS adds to cloud
model to outline the vision of “computing utilities" into computing by providing users access to computing
reality. Cloud Computing provides gives a malleable and resources such as virtual machines, systems, servers,
simple approach to store and recover gigantic information database storage.
without stressing over the development maintenance of
computing infrastructure. It provides shared resources, 2) Private Cloud- In contrast to the public cloud type,
information, software and other devices as per the end private cloud is maintained by an individual enterprise.
client's prerequisite at particular time. It provides Unlike public cloud it is committed to a solitary association
inescapable, advantageous , on-request organize access to a
common pool of configurable computing resources such
as virtual machines, networks, servers, storage, applications

ISBN: 978-1-5386-4119-4/18/$31.00 ©2018 IEEE 424

Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on May 03,2024 at 12:20:03 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Advances in Computing, Communication Control and Networking (ICACCCN2018)

For example- Federal Agency Some of the dynamic load balancing algorithms is as
follows:
a) Game Theory Load Balancing Algorithm
3) Hybrid Cloud- Hybrid Cloud is combination of on-
b) Stochastic Hill Climbing Algorithm
premises private cloud and third-party public cloud c) Genetic Load Balancing Algorithm
services with association between the two. Hybrid clou d
d) Ant Colony Optimization Based Load Balancing
allows migration of workloads amongst private and public
Algorithm
cloud, which brings about more prominent adaptability and
versatility. e) Honey Bee Behavior Inspired Load Balancing
Now a day’s demand for cloud computing resources has Algorithm
f) Fireflies Based Load Balancing Algorithm
been increasing and in this manner number of end clients
g) Particle Swarm Optimization Based Load Balancing
getting to those services is expanding. Because of which
need to adjust the load of the distributed servers emerges . Algorithm
Load Balancing refers to dissemination of network or
application traffic across a pool of servers more equally.
Load balancing enhance performance and reliability 2) Process initiated by whom- On the basis of process
of applications. Load balancing contributes to cloud initiation, load balancing algorithms can be divided into:-
computing by providing an effective solution for various Sender Initiated
issues residing in cloud computing environment set-up Receiver Initiated
and usage. Load balancing focus on resource allocation Symmetrically Initiated
and task scheduling in distributed environment. It also
helps in avoiding bottlenecks issue in network traffic, The rest of the paper is organized as follows: section 2 deals
reducing response time, lessening overall cost, with related work. Section 3 covers one of the heuristics
enhancing adaptation to non-critical failure capacity algorithms named as Honey Bee Behavior inspired Load
and comprehending security issue. Balancing. The existing system, its model framework, the
Load Balancing algorithms are classified on the basis of advantages and disadvantages of the existing system is also
two factors: discussed. It deals with the new system built using the
1) State of System- Load balancing algorithms further can proposed Ant Colony Optimization, its model framework
be divided into two categories depending on the current and most importantly, its advantages over the previous
state of the system. They are as follows: system. Section 4 includes the results analysis of the two
Static Load Balancing- In static algorithms, approaches. Finally, section 5includes the concluding
information remarks, the direction and scope of future work.
about the system is already known which
incorporates processing power, memory, execution and II. RELATED WORK
information about client's necessities. These algorithms Information Technology is moving to cloud day by d ay.
needn’t bother with the data in regards to current state of Basically, cloud computing is a sort of outsourcing of PC
the framework. This sort of algorithms have critical programs. Cloud computing enables clients to access
issues if there is an occurrence of sudden collapse of computing resources from wherever they require. As it is
system asset, task and furthermore task can't be moved facilitated by third party in the cloud, so end users don't
during its execution for load balancing. Some of the static need to stress over things, like database storage and power.
load balancing algorithms is as follows: In this paper [1], authors analyzed the idea of cloud
a) Round Robin Load Balancing Algorithm computing and discussed why it came into picture. Author
b) Min-Min Load Balancing Algorithm discussed the architecture, cloud computing services and
c) Max-Min Load Balancing Algorithm deployment models of cloud computing. In this paper, they
d) Opportunistic Load Balancing Algorithm discussed the cloud key security challenges and security
e) Throttled Load Balancing Algorithm issues of cloud computing as a major concern.
In paper [2] T. Deepa and co-author gave an outline of load
Dynamic Load Balancing- Any prior information about balancing in cloud computing. The paper also demonstrates
the system is not required. Dynamic algorithms are classification of load balancing algorithms. In this paper,
decision concerning load balancing based upon the current merits and demerits for multiple load balancing algorithms
state of the system. The dynamic algorithms are mind are discussed.
boggling, yet they can give better execution and adaptation And in [3] survey for various load balancing algorithms in
to non-critical failure. Dynamic load balancing algorithms cloud computing has been provided. The paper likewise
are further classified into- gives insights about different parameters, and on the basis of
Distributed Algorithms- Further it is categorized into- those parameters, comparison has been provided for the
Cooperative Type already proposed algorithms.
Non-Cooperative In this paper [4], an algorithm is planned for load counter
Non- Distributed Algorithms- Further it is categorized balance in cloud computing environment in light of conduct of
into- Centralized and Semi-Distributed honey bee food reaping mechanism. The proposed algorithm
counterbalances the load in cloud and enhances the overall
performance of cloud system. This algorithm deals

425
Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on May 03,2024 at 12:20:03 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Advances in Computing, Communication Control and Networking (ICACCCN2018)

Create data center, data brokers,

with execution time, response time as the major QoS
host, virtual machine, cloudlets
parameter. Later on, the algorithm will be improved by
inspect different QoS parameters and furthermore
differentiate the consequence of the proposed calculation
with new algorithm. In this paper [5], authors have talked
about on fundamental aspects of Cloud Computing and Identify overall load
Load balancing. In expansion to that, they have discussed
the load balancing approach that depends on Swarm
Intelligence. In this paper, a heuristic approach based on
ACO has been proposed to begin the load distribution
under cloud computing engineering. The pheromone
amend procedure has been demonstrated as a productive Assign priority to all jobs
tool to adjust the load. This algorithm aims at minimizing
make span of the cloud service. This strategy does not
consider the fault resilience issues. So we can continue to
incorporate the adaptation to internal failure issues in our
future Result Evaluation
explores.
Performance Metrics
Existing Model and limitations
There has been various performance parameters on the
basis of which efficiency of proposed load balancing Scheduling of tasks in cloud computing is an NP-
algorithms are measured. Those metrics are as follows: hard optimization problem. Load balancing of non-pre-
Overhead: Overhead represents the extra cost needed for emptive independent tasks on virtual machines (VMs) is a
implementation of load balancing algorithms. To enhance critical part of task scheduling in clouds. . At whatever point
the efficiency of algorithm, overhead should be low as certain VMs are over-loaded and remaining VMs are
much as possible. under loaded with tasks for processing, the load has to
Throughput: Throughput is used to assess the total be must be adjusted to accomplish ideal machine
number of tasks, which has been finished effectively. High usage. So, an algorithm was proposed named honey bee
throughput enhances overall system performance. behavior inspired load balancing (HBB-LB), which
Fault Tolerance: It represents the capacity of an algorithm expects to accomplish all around adjusted load across
to perform load balancing in case of any failure in any over virtual machines for expanding the throughput.
hubs of the network. The proposed approach additionally balances the task
Migration Time: It indicates the total time priorities on the machines such that the measure of
required to transfer the resources from one hub then onto waiting time of the assignments in the queue is negligible.
the next. To add on to the efficiency of algorithm it should In this paper [6], Authors proposed a dynamic load balancing
be minimized. Response Time: It indicates the time taken algorithm named as Honey Bee Behavior Inspired
by an algorithm to respond to a request. It should be Load Balancing Technique. This dynamic strategy
minimized to enhance the overall system performance. won't just balance the load yet additionally consider task
Resource Utilization: It guarantees the best possible usage priorities in the weighting queue of VMs . This approach
of assets, which incorporates the entire system. It ought to proposes that load balancing in cloud computing can
be maximized to enhance the efficiency of algorithm. be accomplished by modeling the food reaping
Scalability: It measures ability of an algorithm to mechanism of honey bees. This algorithm is borrowed
perform uniformly even if no of end users increments. from a descriptive study of the conduct that honey bees
An algorithm should be highly scalable. endorse to discover and acquire nourishment. In bee colonies,
Performance: It checks for the effectiveness of the there is a class of honey bees called the scout honey bees
system after performing load balancing. which scavenge for nourishment sources, after
Makespan: It indicates the maximum completion time or discovering one, they return to the bee hives to publicize this
the time when the resources are allocated to a user utilizing a move called waggle/tremble/vibration move.
The exposure of this dance, gives the possibility of the
quality and additionally amount of nourishment and
III. SYSTEM MODEL
furthermore its distance from the bee hives. Forager honey
Flowchart for the system bees at that point take after the Scout Bees to the area
of sustenance and afterward start to harvest it. They at that
point come back to the bee sanctuary and complete a
waggle or tremble or vibration move to different honey
bees in the hive giving a thought of how much
nourishment is left and henceforth bringing about either
more misuse or relinquishment of the sustenance source.
In a similar way, the removed tasks from over loaded VMs
are considered as the honey bees . Upon accommodation to
the under loaded VM, the task will refresh the quantity of
various
426
Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on May 03,2024 at 12:20:03 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Advances in Computing, Communication Control and Networking (ICACCCN2018)

priority tasks and heap of that specific VM to all Process of Load Balancing:
other weighting assignments. This will be useful (1) Compute moving likelihood (probability) for the majority
for other assignments in selecting their virtual machine in of its neighbors and select the greatest one as its next goal;
view of load and priorities. At whatever point a high (2) Now judgment is made that another hub is hopeful hub or
priority assignment must be submitted to different VMs, it not in the wake of moving to it. On the off chance that yes,
ought to consider the VM that has least number of high create a retrogressive subterranean insect and instate this
priority task so the specific task will be executed at the regressive insect. For forward insect, backpedal to stage 1;
most punctual. Since all VMs will be arranged in (3) The regressive insect backpedals to the beginning stage of
increasing request in view of load, the assignment its forward subterranean insect, alongside the way of its
eliminated will be submitted to under loaded VM. forward insect with the other way. Refresh the data
Generally, the tasks are the honey bees and the VMs are pheromone of every hub the regressive subterranean insect
the sustenance sources. Stacking of an assignment to a VM cruises by, and evacuate the retrogressive insect
is like a honey bee hunting a nourishment source (a bloom when achieving the beginning stage;
or a fix of blossoms). At the point when a VM is over- (4) Calculate the total assets of the hopeful hubs, and stop the
burden i.e., like the honey getting drained at a procedure in the event that they can take care of the demand
nourishment source, the task will be scheduled to an of load adjusting and,
under loaded VM like a scrounging honey bee finding (5) Perform the load balancing activity
another sustenance source. In this way load balancing is
performed using SI technique called HBB inspired. Algorithms
Limitations-
Honey bee Behavior propelled load balancing intends Honey Bee Behavior Inspired
to accomplish well balanced load across virtual Equations-
machines for augmenting the throughput yet the selection For Capacity of a Virtual Machine
issue of physical hosts for deploying requested task still Cm= penumj *pemipsj*vmbwj
exists. As HBB just streamlines the applicant target host pe- Processing elements
inside a algorithm cycle and after that select the ideal penumj - no. of processors in VM j
target hosts to accomplish the prompt load balancing pemips -million instructions per second
impact. Be that as it may, the prompt impact doesn't vmbwj -Communication bandwidth utility of VM j
ensure high execution proficiency for the next task in spite Capacity of all VM-
of the fact that it has capacities in accomplishing high C=∑mi=1 Ci
resource usage. Likewise it has a few hindrances like- Lack Load on a VM
(.)
of versatility. Lv,Mi,t=
(,)
File can't be moved in legitimate way. N(T.t) represents no. of tasks at time ton service queue of
Lots of programmers could hack the information. VMi
Improper security. S(Vmi,t) represents service rate of VMi, at time t.
Non-Proficiency. Load of all virtual machines-
Less Scalable L= ∑m i=1 Lvmi
Processing Time of a VM:
Proposed model: 
PTi =

So to accomplish the general load balancing in a long
Processing Time of all VM:
term process rather than the quick load balancing
approaches, a novel heuristic approach that depends PT=

on Ant Colony Optimization (ACO), the most prevalent Standard Deviation of Load:
SI method, in Clouds by considering task priorities. In 
σ =  ∑m i=1 (PTi –PT)2
particular, we define our concern as limiting the 
weighted flow time of a set of tasks, i.e., the weighted Step 1: From the above given equation, find capacity and
sum of job finish times minus job start times, while load for all virtual machines. And check system is balanced
likewise limiting makespan i.e the aggregate execution or not:
time of all tasks. Our scheduler basically incorporates If(σ <= Ts ) where Ts is threshold value set[0-1]
a Cloud-wide VM (Virtual Machine) allocation policy System is balanced.
based on ACO to map VMs to physical hosts, in Exit.
addition to a VM-level strategy for considering
individual tasks priorities that bases on the opposite of Step 2: If system is not balanced, Perform Load Balancing
the outstanding ''convey'' from operating systems If L> maximu m capacity
hypothesis Simulation comes about demonstrate that Load Balancing not possible
contrasted and the current works, the proposed Else
approach has lessened the failure number of task Perform Load Balancing
deployment events obviously, enhanced the
throughput, and advanced the outside
administrations execution of cloud data centers.

427
Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on May 03,2024 at 12:20:03 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Advances in Computing, Communication Control and Networking (ICACCCN2018)

Step 3: On the basis of loads on VMs, Divided them

S.No. Parameters ACO based HBB inspired
into groups as – Low Loaded VM (LVM), Balanced VM
(BVM) and Over-Loaded VM(OVM). 1 Broker cost Lesser than HBB Two times
higher than
Step 4: Compute priority of tasks by considering ACO
QoS parameters like deadline, task length, MIPS value of VM.
Step 5: Load Balancing Process: Supply of each machine in 2 Time duration Lesser than HBB Four times than that
LVM is- Maximum capacity – (L/C) of the time
Demand of each machine in OVM is- (L/C) - Maximum required by ACO
capacity
Step 6: Sort VMs in OVM in descending order 3 Bandwi dt h Great er Lesser
Bandwidt h Bandwidt h
Step 7: Sort VMs in UVM in ascending order
Step 8: If there are more than one VM in UVM 4 Ram speed Four times Less than that
Step 9: VMd=Call Pareto optimal Scheduling than that of of the ACO
the HBB
Step10: Update the no. of tasks assigned to VMd
Step 11: Update sets OVM, BVM and LVM 5 Overall cost Lesser than Two times
that of the higher than
HBB ACO

Ant Colony Optimization Based Algorithm

Inputs are :n = VM count, pheromone[n][n], probability[n].
Output - UVM (under loaded virtual machine) which contains We have also presented graph for Response Time (fig:3)
least number of cloudlets. where X- axis represents no. of cloudlets’ and Y-axis
Steps for ACO based Load Balancing Algorithm: represents scalability. Here blue color represents HBB
Step 1: Initialize pheromones[n][n] on the trail. algorithm and Red color represents ACO. From the above
Step 2: Move ants through VMs to lay pheromone graph we can analyse that ACO needs less response time
Step 3: Calculate probability[n] table to respond to a request while HBB needs higher response
Step 4: Select UVM with highest probability from time.
probability[n]
Step 5: Allocate cloudlet to the selected UVM (i.e perform
load balancing task)

IV. Result Analysis

We have presented bar graphs (fig:1 and fig:2) for comparing
both of the algorithms where x-axis represents cloudlets
and y-axis represents no. of servers. For comparison we
have consider factors like Broker cost, Time duration,
Bandwidth, Ram speed, Overall cost. From the above
bar graphs several analysis are done.

Figure 3: Response Time Graph

We have also presented graph for scalability (fig:4) where X-
axis represents no. of cloudlets’ and Y-axis represents
scalability. Here blue color represents ACO algorithms and Red
color represents. From the above graph we can analyze that
ACO provides higher scalability. Also we can see that after a
certain value, both the lines meet at some point
(approximately 40 numbers of cloudlets), which depicts that
ACO performs better when for small space
Figure 1: Bar Graph for HBB

Figure 2:
Authorized licensed useBar Graph
limited for ACO
to: Lovely Professional University - Phagwara. 428
Downloaded on May 03,2024 at 12:20:03 UTC from IEEE Xplore. Restrictions apply.
Figure 4: S calability Graph
 International Conference on Advances in Computing, Communication Control and Networking (ICACCCN2018)

V. CONCLUS ION Behavior Algorithm in Cloud Computing” in IEEE,

Although there are many existing static load 2016 978-1-5090-0893-3/ 16
balancing techniques, but are not suitable for cloud [5]. Deepak Mahapatra, Gaurav Kumar Saini, Himanshu
environments where load will be varying at varying times. Goyal, Amit Bhati, Dep. Of CS “ ACO- A solution of
So we have implemented Honey bee behavior inspired Load balancing in cloud” in International Journal of
dynamic load balancing (HBB-LB), which aims to Engineering Applied Science and Technology, 2016
achieve well balanced load across virtual machines for Volume 1, Issue 3, ISSN No.2455-2143.
maximizing the throughput. Honey Bee Behavior inspired [6]. L.D. Dhinesh Babu, P.Venkata Krishna, School of
load balancing aims to achieve well balanced load across Infromation and Technology, VIT University,
virtual machines for maximizing the throughput but Vellore India” Honey Bee Behavior Inspired load
the selection problem of physical hosts for deploying balancing of tasks in cloud computing environments
requested tasks still exists. As HBB only optimizes the in Applied Soft Computing 2013.
candidate target hosts within an algorithm cycle and then [7]. Santosh Kumar and R.H Goudar, “Cloud
pick out the optimal target hosts to achieve the immediate Computing: Research Issues, Challenges,
load balancing effect. However, the immediate effect Architecture, Platforms and Applications: A survey
doesn’t guarantee high execution efficiency for the next in International Journal of Future Computer and
task although it has abilities in achieving high resource Communication, Vol. 1 no.4, December 2012.
utilization. Also it had some disadvantages of less [8]. Dharmesh Kashyap, Jaydeep Viradiya “A Survey Of
scalability, less secure and less efficient. Various Load Balancing Algorithms In Cloud
So, to minimize the drawbacks of HBB we used Ant Colony Computing” in International Journal of Science and
Optimization for job allocation process . This load balancing Technology Research Volume 3, Issue 11, November
technique based on Ant Colony Optimization gives optimal 2014
resource utilization. The performance of the system [9]. D. A. Menasce and P. Ngo, “Understanding cloud
is enhanced with high availability of resources, computing: Experimentation and capacity planning,”
thereby increasing the throughput. This increase in in Computer Measurement Group Conference, 2009.
throughput is due to the optimal utilization of resources . The [10].S.-C. Wang, K.-Q. Yan, W.-P. Liao and S.-S. Wang,
proposed approach results in reduction of the failure number “Towards a load balancing in a threelevel cloud
of task deployment events obviously, improvement of computing network,” in Computer Science and
the throughput, and optimization of the external services Information Technology (ICCSIT), 2010 3rd IEEE
performance of cloud data centers . International Conference on, vol. 1, pp. 108–113,
Also we have compared the proposed ACO based system IEEE, 2010.
with HBB based system considering parameters like [11]. T. D. Braun, H. J. Siegel, N. Beck, L. L. B¨ol¨oni, M.
response time and scalability. Maheswaran, A. I. Reuther, J. P. Robertson,M. D.
Future work is to implement this technique with Theys, B. Yao, D. Hensgen, et al., “A comparison of
the consideration of server’s CPU power, memory etc eleven static heuristics for mapping aclass of
while redistributing load and also to do implementatio n of independent tasks onto heterogeneous distributed
algorithm for finding out neighbors of a node with particular computing systems,” Journal of Parallel and
Data Centre Network Architecture. Also for comparison Distributed computing, vol. 61, no. 6, pp. 810–837,
we can consider some other QoS factors. 2001.
[12]. T. Kokilavani and D. Amalarethinam, “Load
REFERENCES balanced min-min algorithm for static meta-task
[1].Akshat A jabrao Uike, M.A.Pund, and Sangram S. scheduling in grid computing.,” International Journal
Dandge, “An overview of cloud computing: of Computer Applications, vol. 20, no. 2,2011.
platforms, security issues and applications” in [13].N. J. Kansal and I. Chana, “Existing load
International Journal of Science Technology balancing techniques in cloud computing: A
Management and Research, Vol. 2, Issue 5, May 2017. systematic review.,” Journal of Information Systems
[2].T.Deepa and Dhanraj Cheelu, “Load Balancing & sCommunication, vol. 3, no. 1, 2012.
Algorithms in cloud computing: A comparative s tudy”
in International Journal of Innovations and
Advancement in Computer Science, Vol. 6, Issue 6,
January 2017.
[3]. Bhatt Hirenkumar.H, Prof. Hitesh, A. Bheda, “An
overview of Load Balancing Techniques in Cloud
Computing Environments” in International Journal of
Engineering and Computer Science, Vol. 4, Issue 1,
January 2015.
[4]. Anureet Kaur Research Scholar , Bikrampal Kaur
Professor” Load Balancing in tasks using honey Bee

429
Authorized licensed use limited to: Lovely Professional University - Phagwara. Downloaded on May 03,2024 at 12:20:03 UTC from IEEE Xplore. Restrictions apply.