Load Balancing in Cloud Computing

Violetta N. Volkova1, Liudmila V. Chernenkaya Elena N. Desyatirikova

Highest School of Cyberphysic Systems and Control Voronezh State Technical University
Institute of Computer Sciences and Technology Voronezh, Russia
Peter the Great St. Petersburg State Polytechnic University [email protected]
St. Petersburg, Russia
1
[email protected]
Almothana Khodar, Alkaadi Osama1
Voronezh State University
Moussa Hajali Voronezh, Russia
1
Faculty of Informatics, Damascus University [email protected]
Damascus, Syria
[email protected]

Abstract—Cloud computing is the emerging internet based nodes. Load balancing helps in the fair allocation of computing
technology which emphasizes commercial computing. Load resources to achieve a high level of user satisfaction and proper
balancing helps in improving the performance of the centralized use of resources. High resource utilization and proper load
server. In the present work, various algorithms are analyzed balancing help minimize resource consumption. This helps to
using an analysis tool, namely, cloud analyst. Comparison is also implement fault tolerance, scalability and avoid difficulties [2].
made for algorithms load balancing.
Load balancing is a method that has helped networks and
Keywords—cloud computing; load balancing; cloud analyst resources, to provide maximum throughput with minimal
simulation; round robin algorithm; AMLB algorithm; throttled response time. Load balancing is performed at two levels in
load balancing algorithm cloud computing [3]:
I. INTRODUCTION • The level of the virtual machine, the mapping is made
Cloud computing is the concept of a "cloud of computing", between applications that are loaded in the cloud on the
according to which programs are launched and output the virtual machine. The load balancer assigns the
results of work in a standard web browser window on a local requested virtual machine to physical computers, which
PC, with all applications and their data necessary for operation balances the load of multiple applications from the PC.
located on a remote server on the Internet. The advantages of • A host level, a mapping between the virtual machine
cloud computing include the following: reduced requirements and host resources that allow processing of several
for the computing power of PCs, increased fault tolerance and incoming application requests.
security, the speed of data processing increases many times,
costs for hardware and software, for maintenance, power are III. EXISTING LOAD BALANCING POLICIES
reduced, and disk space is saved. There are various load balancing algorithms used in cloud
Cloud computing is a technology that helps to exchange computing. In this study, the following three algorithms have
data and provide a lot of resources to users. Users pay only for been studied, which can be implemented in the Cloud Analyst
the resources that they used. Cloud computing stores data and simulator [4].
distributed resources in an open environment, and the amount A. Round Robin Algorithm (RR)
of data storage increases very quickly. Thus, load balancing is
the main task in the cloud environment. Load balancing helps This is the simplest algorithm that uses the concept of a
to distribute the dynamic workload across multiple nodes to quantum of time or interval (Fig. 1).
ensure that no node is overloaded.
This study mainly focuses on analyzing the performance of
cloud computing and comparing various load balancing
algorithms using the Cloud Analyst network simulator.
II. LOAD BALANCING
Load balancing is used to distribute more load to smaller
processing nodes to improve overall system performance [1].
In a cloud computing environment, load balancing needs to
distribute the dynamic local workload evenly between all Fig. 1. Round Robin Algorithm (RR).

387 978-1-5386-4340-2/18/$31.00 ©2018 IEEE

 Here time is divided into several sectors, and each node is When processing a client request, if the corresponding VM
given a specific time quantum or time interval, and in this is not found, the load balancer returns "-1" to the data center.
quantum the node will perform its operations. The center request is processed by the data center.
In Round Robin, scheduling a time quantum plays a very
important role, because if the time slice is very large, then the
Round Robin scheduling algorithm is the same as the FCFS
planning [4].
The disadvantage of the method is that, although the
algorithm is very simple, but to determine the quantum size, it
generates an additional load on the scheduler. In addition, it has
higher context switches that increase the turnaround time, and
low throughput.
B. Active Load Balancing Monitoring (AMLB)
This algorithm has a dynamic character. It stores
information about each VM virtual machine and the number of
requests that are currently assigned to each VM. When the
request is distributed by the new VM and if there are several
VMs, the first recognized one is selected, and the AMLB
returns the VM identifier to the data center controller. The data Fig. 3. Throttled Algorithm (TLB).
center controller warns the AMLB about the new distribution
and sends the request to the virtual machine known under this When processing a client request, if the corresponding VM
VM identifier (Fig. 2). is not found, the load balancer returns "-1" to the data center.
The center request is processed by the data center.
IV. CLOUD ANALYST SIMULATOR
The simulation and analysis of the performance of the three
load balancing algorithms are performed using the "Cloud
Analyst" tool [5]. It allows the user to run multiple simulations
with small parameter changes, and also allows you to
customize the location of the users who create the application
and the location of the data centers [6]. Let's indicate the
terminology of the emulator (Fig. 4):

Fig. 2. AMLB Algorithm (AMLB). - Region: in Cloud Analyst, the world is divided into 6
regions that coincide with the 6 major continents in the world;
The disadvantage of the algorithm is that AMLB always - User Base: User Base is considered as a single unit, and is
finds the least loaded VM to assign a new incoming request, used to generate traffic;
but does not check whether it was used earlier or not (therefore
some VM is used intensively, and some are still not involved). - Data Processing Center: brokerage services determine
which center should accept and process the request that comes
C. Throttled Load Balancing Algorithm (TLB) from each user database;
In this algorithm, the load balancer maintains a table of
virtual machine indexes, as well as their states (Available or - VmLoadBalancer: it is responsible for distributing the
Busy). The client / server first makes a request to the data load to the available data center. VmLoadBalancer distributes
center to find a suitable virtual machine (VM) to perform the the load in the data center based on the load balancing policy.
recommended task (Fig. 3).
The data center requests a load balancer to distribute the
virtual machine. The load balancer scans the index table from
above until the first available virtual machine is found or the
index table is completely scanned.
If a virtual machine is found, the data center passes the
request to the virtual machine identified by the identifier. In
addition, the data center confirms the load balancing of the new
distribution, and the data center appropriately revises the index
table.

Fig. 4. Cloud Analyst Simulator.

388
 In the modeling process, CloudSim 4.0 software was used.
V. SIMULATION AND EXPERIMENT
Simulation and virtual experiment are the best way to test
the algorithm in cloud computing. Consider the work of each
of the three load balancing algorithms using the example of the
social network Facebook, which has more than 200 million
registered users around the world (Table 1).

TABLE I. REGISTERED USERS OF FACEBOOK AROUND THE WORLD

Region ID Region Users
North America 0 80 million Fig. 5. Configuration of the data center.
South America 1 20 million
Europe 2 60 million
Limit the model to the fact that each user database is
contained in one time zone, and assume that most users use the
Asia 3 27 million
application in the evenings after work about 2 hours.
Africa 4 5 million
Oceania 5 8 million Suppose also that 5% of registered users will be on-line at
peak time simultaneously and only one-tenth of this number
will be on the network during off-peak hours. Suppose that
For the modeling, suppose we have a similar system, but on each user makes a new request every 5 minutes, when on-line.
a scale of 1/10. Define 6 user databases representing the above
6 regions, with the following parameters (Table 2). VI. SIMULATION RESULTS
We performed the simulation three times in accordance
TABLE II. USER DATABASE SETTINGS with the previous parameters. Each time, we changed the load
Peak hour Users online Users online balancing algorithm, which was analyzed. The results were
Reg Peak hour compared by the criteria: total response time (Figure 6), data
Base Timezone (Local during peak in non-peak
ion (GMT)
time) hours hours center time (Figure 7), hourly data center load (Figure 8-10),
UB1 0 GMT – 7.00–9.00 13:00- 400,000 40,000 and processing costs (Table 4).
6.00 pm 15:00
UB2 1 GMT – 7.00–9.00 15:00- 100,000 10,000
4.00 pm 17:00
UB3 2 GMT + 7.00–9.00 20:00- 300,000 30,000
1.00 pm 22:00
UB4 3 GMT + 7.00–9.00 01:00- 150,000 15,000
6.00 pm 03:00
UB5 4 GMT + 7.00–9.00 21:00- 50,000 5,000
2.00 pm 23:00
UB6 5 GMT+10 7.00–9.00 09:00- 80,000 8,000
.00 pm 11:00

We also define the data processing center (Table 3), which

must process the request coming from each user database with
the following parameters (Fig. 5). Fig. 6. Total response time.

TABLE III. DATA CENTER SETTINGS

Name Dc1
Region 0
Arch X86
OS Linux
VMM Xen
Cost per VM $/Hr 0.1
Memory Cost $/s 0.05
Storage Cost $/s 0.1
Data Transfer Cost $/GB 0.1 Fig. 7. Data center processing time
Physical Hw Units 20

389
 REFERENCES
From the above review of load balancing and the three
existing policies for the Cloud Analyst simulator, you can
conclude that load balancing is a complex task in cloud
computing.
Comparing the results obtained using different load
balancing algorithms, we can conclude that the overall
response time in the Throttled algorithm is better than in other
algorithms, and the data center time is also better.
Fig. 8. Hourly data center load in (RR).
REFERENCES
[1] Buyya R., Ranjan R. and Calheiros R.N. Modeling and simulation of
scalable cloud computing environments and the cloudsim toolkit:
challenges and opportunities, High Performance Computing &
Simulation HPCS'09, 2009, pp. 1-11.
[2] Desyatirikova E. N., Kuripta O. V. Quality management in IT service
management based on statistical aggregation and decomposition
approach, 2017 International Conference "Quality
Management,Transport and Information Security, Information
Technologies" (IT&QM&IS) , 2017, pp. 500-505. DOI:
10.1109/ITMQIS.2017.8085871.
Fig. 9. Hourly data center load in (AMLB). [3] Calheiros R.N. CloudSim: A Novel Framework for Modeling and
Simulation of Cloud Computing Infrastructures and Services, Eprint:
Australia, 2009, pp.9–17.
[4] Simar P.S. , Anju S. and Rajesh K. Analysis of load balancing
algorithms using cloud analyst, International Journal of Grid and
Distributed Computing, vol. 9, No. 9, 2016, pp.11-24.
[5] Maguluri S.T., Srikant R. and Ying L. Stochastic models of load
balancing and scheduling in cloud computing clusters, in: INFOCOM
Proceedings IEEE, 2012, pp. 702–710.
[6] Bhathiya W.A. CloudAnalyst: A CloudSim based Tool for Modelling
and Analysis of Large Scale Cloud Computing Environments. University
of Melbourne: Australia, 2009, p.21-35.

Fig. 10. Hourly data center load in (TLB).

TABLE IV. COST OF PROCESSING

Algorithm VM Cost , $ Full Value , $
RR 120.5 632.34
AMLB 112.8 580.67
TLB 99.1 501.92

390