Load Balancing Algorithm to Improve Response Time on Cloud Computing

International Journal on Cloud Computing: Services and Architecture (IJCCSA) Vol. 7, No. 6, December 2017
DOI: 10.5121/ijccsa.2017.7601 1
LOAD BALANCING ALGORITHM TO IMPROVE
RESPONSE TIME ON CLOUD COMPUTING
Nguyen Xuan Phi1
and Tran Cong Hung2
1,2
Posts and Telecommunications Institute of Technology, Ho Chi Minh, Vietnam.
ABSTRACT
Load balancing techniques in cloud computing can be applied at different levels. There are two main
levels: load balancing on physical server and load balancing on virtual servers. Load balancing on a
physical server is policy of allocating physical servers to virtual machines. And load balancing on virtual
machines is a policy of allocating resources from physical server to virtual machines for tasks or
applications running on them. Depending on the requests of the user on cloud computing is SaaS (Software
as a Service), PaaS (Platform as a Service) or IaaS (Infrastructure as a Service) that has a proper load
balancing policy. When receiving the task, the cloud data center will have to allocate these tasks efficiently
so that the response time is minimized to avoid congestion. Load balancing should also be performed
between different datacenters in the cloud to ensure minimum transfer time. In this paper, we propose a
virtual machine-level load balancing algorithm that aims to improve the average response time and
average processing time of the system in the cloud environment. The proposed algorithm is compared to the
algorithms of Avoid Deadlocks [5], Maxmin [6], Throttled [8] and the results show that our algorithms
have optimized response times.
KEYWORDS
Load Balancing, Response Time, Cloud Computing, Execution Time.
1.INTRODUCTION
As load balancing algorithms have appeared to respond the cloud computing problems, load
balancing on the cloud have been recently studied and suitable algorithms have been implemented
in this area. In terms of load balancing, we often refer to some basic algorithms such as Round
Robin algorithm, Weighted Roud Robin algorithm, Least Connection algorithm, Weighted Least
Connection algorithm, Least Response Time algorithm. These are algorithms basically, which use
methods such as round robin which choose the machine with the least connection to handle the
task, choose the job has the smallest response time. There are also many other load balancing
algorithms depending on the software or hardware load balancer used. The load balancing
technique in cloud computing can be applied at different levels, depending on what the load
balancing is. The problem is how to distribute those requests so that the response time is minimal
and how to decide the number and characteristics of the virtual machines that handle the requests.
These requests must be distributed to the available virtual machines for processing. Therefore,
how to decide the number and characteristics of the virtual machine to handle these requests, and
how to distribute such requests to optimize the response time, is the task that the solution load
balancing must resolve.
Load balancing policies depend on how we want to perform. For example, if you want the data
center to load physical hosts, ie, the physical host allocation policy for the virtual machine, such
as a host allocation policy that has the least processing core used for a virtual machine. Such load
balancing is called at the Host level. If you want to balance the load of virtual machines running
the application, that is, each virtual machine divides resources received from the host for the task

2
or application service running on them, load balancing at this level is called virtual machine level.
In order to meet the above requirements, the establishment of an efficient load balancing
algorithm and how to use resources in a reasonable manner is the goal that cloud computing
wants to achieve. [1], [2]. Load balancing techniques in cloud computing now look at different
parameters such as performance, response time, scalability, throughput, resource utilization, fault
tolerance, accommodation and related costs. In addition to energy efficiency, carbon emissions
are also considered [3].
With such usability and performance, cloud computing has become an indispensable trend. In the
future, the increase in the number of cloud users requires service providers to meet the needs of
users with minimal response time. Therefore, load balancing methods in cloud computing are
increasingly being developed, when number of servers or server configurations increasing is only
temporary method. Effective use of resources on the "cloud" is a necessity. This is also a huge
challenge in the field of cloud computing. In order to meet the above requirements, the
establishment of an efficient load balancing algorithm and how to use resources in a reasonable
manner is the goal that cloud computing wants to achieve [4], [5], [6], [7], [8].
Response time on cloud computing is very interested in the research. Shubham Sidana [9] et al.
presented the NBST algorithm, balance the load by arranging the virtual machines on the basis of
their processing power and arranging the cloudlets according to their Length i.e. number of
instructions in the cloudlet. The list of virtual machines and cloudlets is then submitted to broker
for the allocation. In the paper [10] Atyaf Dhari et al. proposed Load Balancing Decision
Algorithm (LBDA) to manage and balance the load between the virtual machines in a datacenter
along with reducing the completion time (Makespan) and Response time. The mechanism of
LBDA is based on three stages, first calculates the VM capacity and VM load to categorize the
VMs’ states (Under loaded VM, Balanced VM, High Balance VM, Overloaded). Second,
calculate the time required to execute the task in each VM. Finally, makes a decision to distribute
the tasks among the VMs based on VM state and task time required. The authors compared the
result of proposed LBDA with MaxMin, Shortest Job First and Round Robin. The results showed
that the proposed LBDA is more efficient than the existing algorithms.
Within the scope of this article, we will focus on virtual machine load balancing. The scheduling
policies used are time-share and space-share for virtual machines and tasks. The aim of the paper
is to propose the improvement of the Throttled algorithm [8], based on the research and
evaluation of three Maxmin algorithm [6], Avoiding congestion in load balancing algorithm[6],
Throttled algorithm [8] to improve the response time and average processing time of the load
balancing system in the cloud environment.
The paper is organized into the following sections: Part1 presents introduction to the load
balancing algorithms. Part 2 presents the our proposed algorithm. Part 3 show the simulation
results of the proposed algorithm. Part 4 conclusion.
2. PROPOSED ALGORITHM
The proposed equilibrium algorithm will be based on the basis that the authors [5], [6], [8] have
done to optimize the average processing time and response time of virtual machines. When
studying the load balancing algorithms [5], [6], [8] in the cloud environment, we find that the
above works of the author group are extremely useful. To develop this algorithm, we tried to
learn more and propose a new algorithm. Therefore, if the Cloud Manager can be further
optimized in the load balancing algorithm by adding some parameters such as the expected
completion time of each resource (vm) With lists of requests coming up, the result will be an
optimal algorithm.

3
In the proposed algorithm, we will look at the parameters such as: the list of workloads of the
system (Cloudlet), the queue list has been submitted for each virtual machine, percentage of
utilization of virtual machine (represented by the number of requests for each virtual machine's
requested queue and the expected cost of completing that queue), the expected completion time
when there is a request for each virtual machine. With these parameters, we will select the virtual
machine with the smallest scheduled completion time and the lowest level of usage for assigning
tasks.
2.1. Theoretical basis
To measure the effect of load balancing can be based on many factors, but the most important are
two factors: load and load performance. Load is the CPU queue index and CPU utilization.
Performance is the average response time required by the user. The load balancing algorithm is
based on input parameters such as the configuration of virtual machines, the length of the cloudlet
tasks, the arrival time, the completion time of the tasks, and then the expected completion time. of
each task, expected response time. Response time is the processing time plus the cost of the
request or task transmission time, queued through the network nodes. Expected response time is
calculated according to the following formula [4]:
Expected Response Time = F – A + Tdelay (1)
where:
F: time to complete the task.
A: arrival time of the task.
Tdelay: transfer time of the task.
Because the algorithm that performs load balancing is that of DatacenterBroker, the level of the
algorithm only affects the processing time in a local environment of a data center. Therefore the
communication delay parameter can be omitted, so Tdelay = 0.
Calculate expected task completion time [4]:
If the scheduling policy is Spaceshare-Spaceshare or Timeshare-Spaceshare, then the formula is
defined by the formula (2), (3):
Where capacity is calculated by the formula [4]:
If the scheduling policy is Space share-Timeshare or Timeshare-Timeshare, it is determined by
the formula (4), (5):
Where capacity is calculated by the formula [5]:

4
In formulas (2), (3), (4) and (5):
- eft(p) is the expected completion time of the Cloudlet p.
- est is the arrival time of Cloudlet p.
- rl is the total number of instructions the Cloudlet p must execute on a processor.
- capacity is the average processing power (in MIPS) of a core for Cloudlet p.
- ct is the current simulation time.
- cores(p) is the number of cores required by Cloudlet.
- np is the actual number of core that the host is considered.
- cap is the processing power of the core.
The capacity parameter defines the true capacity for processing tasks on each virtual machine.
Clearly, capacity depends on the policy of scheduling computing resources on virtualized
systems. Total processing power on a physical host is constant and depends on the number of
physical cores and processing power of each cores. However, when this processing resource is
shared for multiple tasks (cloudlet) simultaneously, each task requires a certain number of cores
and if the total number of cores is greater than the number of physical cores, the virtual core
concept appears, each virtual core will have lower power processor capabilities of physical cores.
In other words, the capacity of a virtual core for a task is only equal to or less than the physical
core and depends on the resource sharing policy. Capacity is the processing power of a virtual
core. From this analysis and based on the resource sharing policy to develop the formula for
capacity. Resource sharing policy is characterized by the scheduling mechanism in cloud
computing. We have two levels of scheduling: virtual machine scheduling to share physical host
machine resources and task scheduling to share virtual machine resources. There are two
scheduling mechanisms, Timeshared and Spaceshared. In this paper, we will implement
algorithms and simulations based on the Spaceshared - Timeshared policy in turn for virtual
machines and tasks. Therefore, the basis for calculating the proposed algorithm will be based on
formulas (4) and (5).
2.2. Algorithm design
Step 1: Initialize Data center Broker. The status table of the virtual machine and the state
of the existing clouds. At the time of initialization no virtual machines were allocated the
Cloudlet.
Step 2: When there is a request to allocate new virtual machine come Data center Broker,
DatacenterBroker analyzes the status table. Then, calculate the total execution time of all existing
cloudlets in the queue (of each virtual machine) and the expected completion time of the new
cloudlet being prepared for processing. If the virtual machine has the smallest processing time
expected, that machine is chosen to submit the next Cloudlet. If there is more than one, the first
virtual machine is selected.
Step 3: Send the selected virtual machine ID to the Data center Broker then
DatacenterBroker sends the cloudlet to the virtual machine allocate by that ID.
Step 4: Databroker notifies about new allocation and updates to virtual machine and
cloudlet status tables.

5
Step 5: When the virtual machine completes the processing request and DatacenterBroker
receives the Cloudlet response, it will update the Cloudlet's status table as completed and reduce a
Cloudlet in the status table.
Step 6: Go back to Step 2.
Figure 1. Flow chart of our proposed algorithm.
Thus, the difference between the algorithms we propose versus the other algorithms is that: Put
into the expected completion time of each resource (vm) for the cloudlet task in queue task. Based
on this parameter, the algorithm will select the VM with the smallest expected completion time
and the lowest percentage of utilization for task allocation. In the our proposed algorithm, we will
Start
Get task requests/loudlets
Select the optimal VM in the
virtual machine list
Total cost = Calculate total expected
completion time (all task queues and
new tasks)
Total cost = Min
Select the VM
that meet the conditions
- Send requests to the selected VM in the
Datacenter and assigns the status as incomplete.
- Add this task to queue of the selected VM.
- Reduce a Cloudlet in the status table.
End
No
Yes
Yes
No
No request to
processing?

6
also look at specific parameters such as queue list (Cloudlet) of all system, the queue list
submitted for each virtual machine percentage of utilization of virtual machine (expressed by the
number of virtual requests submitted by each virtual machine and the expected cost of completing
that queue), expected completion time when there is a request come to each virtual machine.
3. SIMULATION RESULTS
3.1. Data using simulation
The goal of this simulation is to compare, analyze, evaluate the response time and execution time
of Throttled algorithms [8] and the proposed algorithm. Using CloudSim Cloud Simulator
includes 1 datacenter. System parameter values are given in the Tables 1, Table 2, Table 3. The
simulation scenarios implemented according to the VM scheduling policy and task policy is:
SpaceShared-SpaceShared.
Table 1: Value of parameters in cloud setting.
Type Parameters Value
Datacenter
Number of datacenter 1
Number of host 3
Host
Number of PEs per host 1-4
MIPS of PE 1000-30000 MIPS
Memory of host 5120-10240-12288
Storage capacity
1024000-1044480MB
(1000-1020GB)
Bandwidth 10000MB
VM
Total Virtual Machine 3
Virtual Machine Memory (RAM) 1024-3072
Bandwidth 1024MB
Cloudlet/task
Total Cloudlet 10-60
The length of the cloudlet 1024-20480
Required PE number 1-3

7
Table2.Parameters of virtual machine
Table 3. Parameters of cloudlets
ID
Cloudlets
Leng of
Cloudlet
Number of
required PE
ID
Cloudlets
Leng of
Cloudlet
Number of
required PE
0 2000 1 15 2000 1
1 3000 2 16 1000 2
2 4000 1 17 3000 1
3 3000 2 18 5000 2
4 2000 2 19 2000 1
5 2000 1 20 2000 1
6 1000 2 21 3000 2
7 3000 1 22 4000 1
8 5000 2 23 3000 2
9 2000 1 24 2000 2
10 2000 1 25 2000 1
11 3000 2 26 1000 2
12 4000 1 27 3000 1
13 3000 2 28 5000 2
14 2000 2 29 2000 1
3.2. Simulation results
In this experiment, we simulate a cloud of the following parameters: 30 cloudlet (task), 1
datacenter, 3 VM; with parameters in Table 1, Table 2, Table 3. Simulation on the Throttled [8]
and our propose algorithm, response time results as follows:
ID
Memory
(Mb)
Bandwith
(Mb)
Number of PE/core Speed of PE (MIPS)
0 4069 1024 2 200000
1 2048 1024 1 100000
2 1024 1024 2 50000

8
Table 4. Results of throttled algorithm [8].
Cloudlet ID
VM
ID
Time
(ms)
Start
(ms)
Finish
(ms)
0 0 125 100 225
3 0 187.5 225 412.5
1 1 422.5 100 522.5
4 0 220 412.5 632.5
6 0 140 632.5 772.5
5 1 250 522.5 772.5
7 0 187.5 772.5 960
9 0 125 960 1085
2 2 1095 100 1195
10 0 220 1085 1305
8 1 642.5 772.5 1415
12 0 250 1305 1555
14 0 125 1555 1680
13 1 375 1415 1790
15 0 220 1680 1900
16 1 220 1790 2010
11 2 815 1195 2010
17 0 220 1900 2120
20 0 125 2120 2245
21 0 187.5 2245 2432.5
19 2 532.5 2010 2542.5
18 1 642.5 2010 2652.5
22 0 330 2432.5 2762.5
25 0 125 2762.5 2887.5
26 0 110 2887.5 2997.5

9
24 1 345 2652.5 2997.5
27 0 187.5 2997.5 3185
23 2 750 2542.5 3292.5
29 0 217.5 3185 3402.5
31 0 187.5 3402.5 3590
Calculate average execution time and average response time of all tasks (cloudlets):
- Average execution time: 319.3333 (ms).
- Average response time: 1911.667 (ms).
Table 5. Results of propose algorithm
Cloudlet ID
VM
ID
Time
(ms)
Start
(ms)
Finish
(ms)
0 0 125 100 225
1 0 235 100 235
3 0 220 225 445
4 0 220 335 555
5 0 220 445 665
6 0 110 555 665
2 1 565 100 665
7 0 187.5 665 852.5
9 1 297.5 665 962.5
8 0 407 665 1072
10 1 330 852 1182
12 0 250 1072.5 1322.5
14 0 125 1322.5 1447.5
13 1 375 1182.5 1557.5
15 0 220 1447.5 1667.5
16 1 220 1557.5 1777.5
11 2 815 962.5 1777.5
17 0 220 1667.5 1887.5

10
18 0 312.5 1777.5 2090
20 1 312.5 1887.5 2200
21 0 220 2090 2310
19 2 532.5 1777.5 2310
23 0 187.5 2310 2497.5
25 0 125 2497.5 2622.5
26 0 110 2622.5 2732.5
22 1 532.5 2200 2732.5
24 2 532.5 2310 2842.5
27 0 220 2732.5 2952.5
30 0 125 2952.5 3077.5
31 0 187.5 3077.5 3265
Calculate average execution time and average response time of all tasks (cloudlets):
- Average execution time: 284.65 (ms).
- Average response time: 1686.467 (ms).
The above results show that our proposed algorithm has better response times than the Throttled
algorithm [8], as shown in Fig 2, Fig 3.
Figure 2. Comparison of execution time (ms): Throttled Algorithm [8] and Proposed Algorithm
Throttled Algorithm Propose Algorithm

11
Figure 3. Comparison of response time (ms): Throttled Algorithm [8] and Proposed Algorithm
From simulation results and comparison charts between average excution time and average
response time of Throttled algorithms [8] and our proposed algorithm, we see that average
excution time and average response time of our proposed algorithm has been improved than
Throttled algorithm [8].
4. CONCLUSION
Our algorithm is proposed from the Throttled algorithm [8]. In the Throttled algorithm [8], the
authors pay attention to the amount of load that virtual machines are making. In the proposed
algorithm, in addition to concentrating on the load, the researcher is able to perform the tasks /
requirements of the virtual machine. In the cloud environment, the distribution of load between
virtual machines is heterogeneous in terms of processing power, so that each virtual machine can
have different processing time costs. For efficient load balancing, choose which virtual machines
cost the least processing time to assign tasks. Our proposed algorithm was improved and inherited
from the throttled algorithm [8] and was tested in the Cloudsim cloud computing environment
and used in the Java programming language. In this article we use the same schedule as
Spaceshared - Timeshared with virtual machines and tasks. From Figures 2 and 3 we find that the
response time and average processing time of the algorithm are significantly improved compared
to the Throttled algorithm. In the future, we will consider consider the security of the load on
cloud computing.
REFERENCES
[1] Jasmin James, Dr. Bhupendra Verma (2012), “Efficient vm load balancing algorithm for a cloud
computing environment”, International Journal on Computer Science and Engineering (IJCSE), pp.
1658-1663.
[2] Hiren H. Bhatt and Hitesh A. Bheda (2015), “Enhance Load Balancing using Flexible LoadSharing in
Cloud Computing”, International Conference on Next Generation Computing Technologies (NGCT),
pp.72-76.
[3] Mayanka Katyal, Atul Mishra (2013),“A Comparative Study of Load Balancing Algorithms in Cloud
Computing Environment”, International Journal of Distributed and Cloud Computing, Volume 1 Issue
2, pp.5-14.
Throttled Algorithm Proposed Algorithm

12
[4] Rodrigo N. Calheiros, Rajiv Ranjan, Anton Beloglazov ,C´ esar A. F. De Rose and Rajkumar Buyya
(2010), “CloudSim: a toolkit for modeling and simulation of cloud computing environments and
evaluation of resource provisioning algorithms”, Software: Practice and Experience (SPE), Volume 41
Number 1, pp.23-50.
[5] Rashmi. K. S, Suma. V, Vaidehi. M (June 2012), “Enhanced Load Balancing Approach to Avoid
Deadlocks in Cloud”, Special Issue of International Journal of Computer Applications on Advanced
Computing and Communication Technologies for HPC Applications – ACCTHPCA, pp.31-35.
[6] Rajwinder Kaur, Pawan Luthra (2014), “Load Balancing in Cloud System using Max Min and Min
Min Algorithm”, International Journal of Computer Applications Proceedings on National Conference
on Emerging Trends in Computer Technology (NCETCT- Number 1), pp.31-34.
[7] Navtej Singh Ghumman, Rajwinder Kaur (2015),“Dynamic Combination of Improved Max-Min and
Ant Colony Algorithm for Load Balancing in Cloud System”, 6th International Conference on
Computing, Communication and Networking Technologies (ICCCNT).
[8] Hafiz Jabr Younis (2015), “Efficient Load Balancing Algorithm in Cloud Computing”, Islamic
University Gaza Deanery of Post Graduate Studies Faculty Of Information Technology.
[9] Shubham Sidana, Neha Tiwari (2016), “NBST Algorithm: A load balancing algorithm in cloud
computing”, International Conference on Computing, Communication and Automation (ICCCA),
pp. 1178 – 1181, IEEE Conference Publications.
[10]Atyaf Dhari and Khaldun I. Arif (2017), “An Efcient Load Balancing Scheme for Cloud Computing”,
Indian Journal of Science and Technology, Vol 10 (11), IEEE Conference Publications.
Authors
Tran Cong Hung was born in Vietnam in 1961.
He received the B.E in electronic and Telecommunication engineering with first
class honors from HOCHIMINH University of technology in Vietnam, 1987. He
received the B.E in informatics and computer engineering from HOCHIMINH
University of technology in Vietnam, 1995. He received the master of engineering
degree in telecommunications engineering course from postgraduate department
Hanoi University of technology in Vietnam, 1998. He received Ph.D at Hanoi
University of technology in Vietnam, 2004. His main research areas are B – ISDN
performance parameters and measuring methods, QoS in high speed networks,
MPLS. He is, currently, Associate Professor PhD. of Faculty of Information
Technology II, Posts and Telecoms Institute of Technology in HOCHIMINH,
Vietnam.
Nguyen Xuan Phi was born in Vietnam in 1980.
He received Master in Posts & Telecommunications Institute of Technology in Ho
Chi Minh, Vietnam, 2012, major in Networking and Data Transmission. Currently
he is a PhD candidate in Information System from Post & Telecommunications
Institute of Technology, Vietnam. He is working at the Information Technology
Center of AGRIBANK in Ho Chi Minh city, Vietnam. His main research areas are
load balancing on cloud computing, optimizing the performance of cloud computing.

Load Balancing Algorithm to Improve Response Time on Cloud Computing

More Related Content

What's hot (20)

Similar to Load Balancing Algorithm to Improve Response Time on Cloud Computing (20)

More from neirew J (20)

Recently uploaded (20)

Load Balancing Algorithm to Improve Response Time on Cloud Computing