Escolar Documentos
Profissional Documentos
Cultura Documentos
Abstract-Cloud computing services are growing very fast Load Balancing strives to distribute the load in equal
especially with the high demand of mobile and online proportions across resources depending on recourse capacity so
applications (Apps) and services. This exponential growth that each resource is not overload or underutilized in a cloud
emphasis on the need of minimizing the makespan scheduling
system [4]. Moreover, it is considered to be a NP-hard
and utilizing the resources efficiently based on dynamic
optimization problem [5] that has 2 types: dynamic load
environment. Accordingly, many load balancing algorithms have
been developed to overcome these issues using intelligent
balancing and Static load balancing. Dynamic load balancing is
optimization methodologies, such as Genetic Algorithms (GA), to rebalance the system while running when detecting
Ant Colony optimization (ACO), Artificial Bee Colony (ABC) overloaded VMs while static load balancing is to balance the
and Particle Swarm Optimization (PSO). This paper surveys the system from the start by scheduling. Static load balancing is
above intelligent optimization techniques and focuses on the Ant preferable as it avoids VM migration costs and provides better
Lion Optimizer (ALO) intelligent technique, also it proposes an execution time and quality of service (QoS) [2].
implementation of ALO based cloud computing environment as
efficient algorithm that expected to supplies better outcomes in
To work on Load Balancing, traditional algorithms like
load balancing.
First Come First Serve (FCFS), Round Robin (RR), Random
Keywords-Cloud computing; Load balancing; Swarm
Allocation (RA), Shortest Job First (SJF) and Longest Job First
intelligence; Ant Lion Optimizer (LJF) aren't sufficient and the meta-heuristic algorithms like
Evolutionary Algorithms and Swarm Intelligent Algorithms
I. INTRODUCTION were explored. Additionally, Swarm Intelligent Algorithms are
Cloud computing has emerged from the need to outsource known for their self organizing properties thus can work well
computing and storage facilities to clients. Its rapid growth is in a dynamic scalable environment like cloud systems [1, 6].
because of the huge enhancement of communication Moreover, As New Swarm Intelligent Algorithms emerge like
technologies and virtualization technologies that lead most of Grey Wolf Optimizer (GWO), Firefly Algorithm (FA) and Ant
its systems nowadays to become essential in day to day Lion optimizer (ALO), so is the need to explore these
activities [1]. Moreover, it delivers reusable efficient Algorithms in cloud computing depending on No Free Lunch
distributed resources to clients in 3 forms of services: theorem [7].
infrastructure as a service (IaaS), platform as a service (PaaS)
and software as service (SaaS). With companies like Amazon, The rest of paper is organized as follows: Section 2 presents
Microsoft, IBM , Google and other companies giving cloud a survey of the related work in Load balancing, While Section
services ,regular clients who has no knowledge of cloud 3 the surveyed algorithms is displayed further, In Section 4 the
building can easily work on what they need [1]. There are details of the ALO Algorithm is explained, Section 5 shows the
many examples of IaaS like Amazon Elastic Compute Cloud proposed work that utilizes the ALO algorithm, and Section 6
(Amazon EC2), Windows Azure, Google Compute Engine and illustrates the conclusion and future work.
RESERVOIR [2][3]; of PaaS like Force.com and Google App
Engine[3] and of SaaS like Google Apps, Office 365 and
II. RELATED WORK
Confluence Cloud. With the increased need for the services of
cloud computing systems, the challenge to enhance Load Balancing is one of the main issues in cloud
performance and quality of service is getting higher, leading computing, as it enhances system response time, power
many researchers to focus on optimization Cloud Computing consumption and utilizes resources efficiently. Many
elements like load balancing [4]. traditional algorithms have been used to optimize the load
210
20J 5 iEEE Seventh International Conference on Intelligent Computing and Information Systems (ICiCIS 'J 5)
balancing for task scheduling. FCFS, RR, RA, SJF and LJF are
examples of these traditional algorithms. These algorithms To enhance Load balancing, researches inspected Artificial
heavily cause the overloading virtual machines (VM) problem, Bee Colony (ABC). ABC depend on 3 types of bees,
which in tum needs a VM migration process as a solution. employed, onlooker and Scout bees to reach optimum food
They may be fast at scheduling but they don't utilize resources sources. The employed bees randomly search for food source
efficiently, the tasks take a long time and it is tough to handle then onlooker bees work on the best of food sources. After a
the dynamic behavior and scalability of cloud computing. while scout bees explore new area in search space. In[l]
Accordingly, many intelligent algorithms have been devoted proposed a task Scheduler based on ABC. However, [11]
such as GA, ACO, ABC and PSO. proposed an improved ABC-based Algorithm called PS-ABC,
which minimize the power consumption and maximize the
Earlier researches have approached Genetic Algorithms performance in VM migration. According to [6] and upon
(GA) for its simplicity in solving optimization problems like comparison ABC results in [1] to ACO in [10], ABC gives
load balancing. The Algorithm depends on Evolutionary better performance than ACO. Also according to the work in
mechanisms. It implements filtration of popUlation by [1, 11, 6] and [12], ABC performs better than GA but suffers
removing the weak individuals and adding new ones created from replication and network overhead.
from crossing over or mutation for the strong individuals. It
was used for a Task Scheduler in [4] and VM Scheduler in [8]. Other researches worked with Particle Swarm Optimization
The work done in [4, 8, 9] indicates that GA gives better (PSO). PSO depends on random movement of particles. In
performance than traditional algorithms. However, it is each iteration, the best position for each particle is saved and
outperformed by other swarm intelligent Algorithms as the best particle is chosen. The algorithm is simple thus is fast
Artificial Bee Colony in [6] and Particle Swarm Optimization and has no network overhead, although it falls for local optima.
in [2] and [7]. To enhance its flaw, work in [5] used an Improved PSO for
task scheduling for minimizing the execution time. Moreover,
Many researchers explored swarm intelligent Algorithms in [13] a mix of PSO and CS (Cuckoo Search) in Task
like Ant Colony Optimization (ACO). The ACO uses ants to Scheduling is proposed, however, it works on homogenous
search for food source randomly, they lay pheromones on path; VMs only. In [2] authors use PSO only to reschedule the extra
the path with highest pheromones is best solution. Using ACO tasks of the overloaded VMs instead of the entire VM
in [10] shows that ACO performs better than GA in [4] by migration. In [14], they summarize the modified algorithms of
comparing both to FCFS, but it suffers from network overhead PSO and introduce an IPSO using Mutation and crossover
and may fall into local optima. However, work in [6] shows mechanisms. The Modifications perform better than PSO, but
that ACO and GA mixed with other Algorithm performance many researches still inspect other Algorithms for better load
nearly equal on the four different network topologies. balancing either static or dynamic.
The work done by these four intelligent algorithms is
TABLE! WORK IN LOAD BALANCING demonstrated further in Table I.
Method Objective Advantages disadvantages
Hybrid task
Need to explore the scheduling
Scheduling PSO It performance better than
Reduce Makespan algorithm with larger Scale in
with cuckoo search in ordinary PSO
cloud computing.
[13]
Artificial Bee Colony
Efficient with scalability ,
with Longest Job Network overhead and need to
Reduce Makespan performance better than SJF,
First Scheduling in explore heterogeneous tasks.
LSF, FCFS and ACO
[1]
Dynamic Load Balancing by
Extra tasks Better solution to VM migration,
moving tasks form overloaded Falling for PSO problems like
Scheduling using give better performance than
VMs by minimizing execution local optima.
PS. In [2] GA.
and transfer time
Trade off between power saving
Hybrid of Bayes Maximize performance and Gives better performance than
and performance may lead to
theorem - ABC for minimize power consumption RA and GA in Migration and
penalty cost due to SLA
VM scheduler in [11] in VM migration power consumption
violation.
better avoidance of local optima
Improved PSO task
Minimize the response time than PSO and give better results Lack of reliability
Scheduler in [5]
than GA and PSO
ACO Task Scheduler The makespan is better than ACO overhead and falling for
Minimizing execution time
in [10] FCFS and RR local optima
Gives better performance and
may fall in local optima also
GA Task Scheduler QoS than FCFS, RR and
Minimizing the makespan ACO and ABC performance
in [4] Stochastic Hill Climbing in
better
various cloud systems.
Cost effective and better
Minimize or avoid self-
GA VM resources utilization than Least-load Need to check on a large scale to
adapting distribution of load
Scheduler in [8] scheduling and Rotating load balance.
by VM migration 2 scheduling
211
20J 5 iEEE Seventh International Conference on Intelligent Computing and Information Systems (ICiCIS 'J 5)
212
20J 5 iEEE Seventh International Conference on Intelligent Computing and Information Systems (ICiCIS 'J 5)
213
20J 5 iEEE Seventh International Conference on Intelligent Computing and Information Systems (ICiCIS 'J 5)
Where LBf is the lower bound around ant lionj and ALj is the storage, software, infrastructure and software development.
position of ant lionj. Many software applications have been used on daily basis even
by regular users, such as Google Apps and Microsoft office
In each iteration, the ALO algorithm calculates the new 365. Thus it is essential to enhance the cloud computing
position of anti by the average of its scaled random walks performance and reliability continuously. One of the key points
towards the corresponding ant lionj and towards the elite ant of performance is load balancing. A load balance in cloud
lion, that is responsible for the best solution in this iteration, computing saves time as it overcomes the overloading of the
using equation (II). VMs, utilizes the resources efficiently and provides the users
with better quality of services. Accordingly, Researchers
Anti explored many algorithms for load balancing from traditional
k k )(UBk - LBk) + LBk LBk (e.g. FCFS) to heuristic (e.g. RR, min-min) and Meta heuristic
_ (NRW.!,] + NRW!,e ]
+ e
(11 ) (e.g. GA, PSO). Hence, as shown in Fig. 6, the ALO is utilized
2 in the task scheduling phase to manipulate scheduling problem
through VMs in cloud environment.
Where NRWi is the normalized random walk of anti
towards the selected ant lionj. Moreover, NRWie is the As represented in Fig. 6, it obtains 3 phases: task sending,
normalized random walk of anti towards the elite ant lion. The scheduler and VM processing. Every cloud user submits tasks
algorithm ALO can be described further in Fig. 5. through the net by any device. The tasks will go to the task
scheduler, which in turn utilizes the ALO, to be assigned to the
appropriate VMs then, after processing, the results return back
to the users.
Initialize the position of n ants and n ant lions randomly,
where n is number of search agents. In task sending phase, each user can produce any number of
Calculate the fitness of ant lions. tasks form any device. Every task has different number of
Find the best fit ant lion and assume it as the elite. instructions, data size and location. Users send tasks to the
While (the end criterion is not satisfied) cloud system as input to the task scheduler.
Update search boundaries LB, UB using Eq.(9)
For every ant The Task Scheduler phase considered as the key to the load
Select an ant lion using Roulette wheel on balancing, ALO will handle task scheduling using cost function
ant lion fitness. that depends on minimizing the makespan and the task total
Create normalized random walk towards time, so it is calculated as follows:
selected ant lion and elite using Eq. (7) and
Eq. (8) Minimize ( Max(FT), TT )
k
(12)
Update the position of ant using Eq. (11)
Calculate ant fitness. Where Max(FT) is the makespan that is calculated by the
End for following equations:
Max(FT) (13)
Replace an ant lion position with its corresponding itij = iti-lj + PTi
ant position it if becomes fitter PTi = NITi x V MSj
Update elite if an ant lion becomes fitter than the elite
End while
FT is a matrix mxn of the tasks' finishing time , ft i j is the
Return Elite
finishing time of task i in VM j ,PTi is the processing time of
taski, NITi is the number of instructions of taski' VMSj is the
Fig. 5.The pseudo code of ALO
speed of VMj.
V. PROPOSED WORK
o task
Avgerage : 3065.6 & minmum : 2605 [I] B. K ruekaew and W. Kimpan, "Virtual Machine Scheduling
(ms) Management on Cloud Computing Using Artificial Bee Colony,"
4500 Proceedings of the International MultiConference of Engineers and
Computer Scientists ,vol. 1,2014.
4000
[2] F. Ramezani, J. Lu, and F. K. Hussain, "Task-Based System Load
3500
Balancing in Cloud Computing Using Particle Swarm Optimization," Int
3000 J Parallel Prog International Journal of Parallel Programming, vol. 42,
2500 no. 5,pp. 739-754,2013.
215
20J 5 iEEE Seventh International Conference on Intelligent Computing and Information Systems (ICiCIS 'J 5)
Conference on Computer Engineering & Systems (ICCES), vol. 12, no. [13] A. Al - maamari and FA Omara, "Task Scheduling using Hybrid
2,pp. 64--69,2015. Algorithm in Cloud Computing Environments," IOSR Journal of
[II] G. Xu, Y. Ding, J. Zhao, L. Hu, and X. Fu, "A Novel Artificial Bee Computer Engineering,vol. 17,no. 3,pp. 96-106,2015.
Colony Approach of Live Virtual Machine Migration Policy Using [14] G. Kaur and S. Sharma, "Research Paper on Optimized Utilization of
Bayes Theorem," The Scientific World Journal,pp. 1-13,2013. Resources Using PSO and Improved Particle Swarm Optimization
[12] U.Singal and S. jain, "An Analysis of Swarm Intelligence based Load (IPSO) Algorithms in Cloud Computing," International Journal of
Balancing Algorithms in a Cloud Computing Environment," Advanced Research in Computer Science & Technology (IJARCST
International Journal of Hybrid Information Technology, vol. 8, no. I, 2014),vol. 2,no. 3,2014.
pp. 249-256,2015.
216