1.

INTRODUCTION
The advancement in Information and technology have paved way for a significant
paradigm shift in technology being offered as service. The computational complexity
have been reduced notably in the recent decade meanwhile, the storage capacity have
been dramatically increased. One of the major hindrances for the business organizations
to expand its ventures is the inability to meet the growing demands of storage and
computational resources within the organization. Cloud computing was introduced in late
1970s as a measure to resolve the issue by providing various services to organizations
through which the overhead of space, resource and complexity was reduced. Although
cloud offers a range of benefits in terms of storage, services and ease of complex
computations, there are also equal hindrances or challenges faced by cloud service
providers (CSPs) and the clients such as Security, efficient techniques for resource
utilization and allocation, Common Standards and effective schedulers for various
tasks /resources requested from the client.

This research is intended to bring out a significant technical contribution in overcoming

the major hindrance of task scheduling in cloud environment. The research aims to bring
out novel algorithms by taking the nature inspired phenomenon as a base. The proposed
research is motivated by various performance studies and the drawbacks in existing
methods. In the following sections, a brief on the Introduction, Important works in
literature, Objective and aim of the research, the phases of research with the problem
analysis, methodologies inducted, results obtained over the experiments and proofs for
the efficiency of the proposed methods with conclusion and future directions are
discussed.

1.1 BACKGROUND
1.1.1 PREFACE TO CLOUD COMPUTING

Cloud computing refers to the different services offered in terms of software, data storage
and networking over the Internet. Thin clients, range of network accesses, shared
resources and increased elasticity are notable characteristics of a cloud technology. Cloud
computing is offered as a range of services namely SaaS which gives software as a
 service, PaaS providing platforms as service and IaaS giving infrastructure as service
respectively [1]. These are deployed through various models such as private, public and
community cloud. The primary services of cloud include data storage and complex
computations. The key advantage of using a cloud platform is that the clients do not
require any computing resources for working with the data and applications as it can be
done over the Internet. Majority of the cloud applications are done to store and process
great volume of data. Cloud services offer a great data storage system that enables the
users to access the data from a centralized storage. The architecture also provides
provision for data upload to the cloud which again can be shared among multiple users
[2]. Some of the key advantage of adapting cloud computing includes low computing
cost, minimized IT infrastructure, less maintenance and ample storage. There are also
equal threats as such as these advantages in terms of confidentiality, protection and
privacy of data. These also have threats from external sources in terms of different
security breaches and attacks.

1.1.2 CLOUD COMPUTING TYPICAL CHARACTERISTICS

Important characteristics of cloud [3] includes,
1. Self centered service on-demand :

This is achieved when a particular customer provides self-computational capability

for example the server time and storage over network as these needs to be done
automatically in the absence of human intervention with the service provider.

2. Larger accessibility to Network :

Provisions are made available over the networks which are accessed with the help of
standard techniques which promote the usage through heterogeneous thin clients such
as mobiles, tablets and local work stations.

3. Pooling of resources:

The service providers’ resources for the computing can be pooled to serve multiple
clients at a single time with the help of a multi-tenant model that has different
resources both physically and virtually and are often assigned / re-assigned on
 demand. There exists an independent scenario when the customer has either the
knowledge or the control on the resources. Typical examples include storage,
processing and sharing of network space.

4. Increased elasticity:

Provisions for performing various tasks shall be assigned and revoked in an elastic
manner and in few cases; it can be done on an automatic manner for dealing with the
supply-demand issue of the resources. As far as the consumers are concerned, these
provisions will appear to be unlimited and can be limited and revoked at any point of
time.

5. Quantized Service:

Cloud System often controls and optimizes the resources in an automatic manner by
balancing the metering capacity in some level of given abstraction which is
appropriate based on the kind of service (e.g., processing, storage and user access
control). The resource usage shall be monitored, reported and controlled for both the
service provider and the consumer.

1.1.3 CLOUD TECHNOLOGIES’ SERVICES

The common services provided by a cloud [4] mainly falls in any one of the models
namely the Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and
Software as a Service (SaaS) as shown in Figure 1.

1. Infrastructure as a Service (IaaS):

IaaS is a model where the infrastructure is offered and managed virtually by the
service providers. With this, the companies would outsource the storage, data center
and networking through the Internet which offers similar services as that of on-
premises infrastructure. Examples include large backups, clustering, networking,
hosting and monitoring. Here, the service provider takes the responsibility of building
the servers and storage and also for providing data center. Some of the IaaS providers
are Amazon EC2, Azure and Google cloud.
2. Platform as a Service (PaaS):

These are built over the IaaS. The cloud service providers deliver the necessary
computing resources which include both the software and hardware components such
as the middleware and OS which are required for the development and testing of
applications. The PaaS enables the cloud users for installing and hosting the data
processing systems and business analytics tools apart from providing necessary
hardware. Some of the leading PaaS providers are Bluemix, AWS, SAP and
CloudBees.

3. Software as a Service (SaaS):

SaaS is designed such that it includes both the Iaas and the PaaS. Here, the service
provider ensures delivery of entire application / software over the cloud as a pay-per-
user basis. It lets the users to easily get access to the software applications such as
mail service over the Internet. Most common such services include MS Office 360,
Adobe cloud, Google GSuite, Salesforce and Oracle CRM.

Figure 1- Cloud Services

1.1.4 CATEGORIES OF CLOUD DEPLOYMENT

There are three modes in which deployment of a cloud could [5-6] be categorized
based on the organizations’ ability for managing and securing assets based on the
business needs.

1. Public cloud:
 These are in general offered as SaaS service for the users over the Internet. It is worth
economical for the users as the service provider themselves would bear the expenses
of necessary bandwidth and required infrastructure. These tend to have limited
configurations and the cost is often estimated based on the usage capacity. These
have limitations such as lack of proper SLAs. Although these serve for purposed such
as reliability and low cost, these are normally not preferred by organizations that deal
with sensitive data which needs strict compliance over security protocols.

2. Private cloud:

As the name indicates, these are used by large individual organizations that build and
process their own data for particular business and IT demands, the service providers
of these clouds have more control on the customizations and scalability aspects. This
kind of infrastructure shall be built on the premises or to an outsourced third party as
a service. In either way, these have the ability to maintain all the hardware and
software over a given network dedicated for a single owner. Many large and medium
scale financial institutions and government organizations would prefer this kind of
technology.

3. Hybrid cloud:

These are combination of both private and public cloud which provides more
flexibility for business operations and has the complete control over the important
operations and assets that are couples with extended flexibility and cost effectiveness.
The hybrid clouds enable the companies to take on advantage over the public cloud
as and when required by mitigating the workload. For example, the business can
either use the public cloud (for email applications) and private cloud for handling
sensitive data and processes particularly during high demand.

1.2 ISSUES IN CLOUD COMPUTING

Cloud Computing relies on connectivity and are based on Internet where the
resources are shared and the applications are accessed over demand [7]. The major
issues to be concerned before adapting to it are as follows.
1. Security:

Cloud services often involve the services of third-party for storage and security
implementation. It is funny to assume that a service provider would strive for the
protecting and securing others’ data when the consumer uses the services at a very
low cost. Security breaches are often forecasted in the cloud environment. These
present a hard threat in the real time.

2. Privacy:

Information pertaining to a user can be accessed by the host without any permission.
The service providers have the provisions to access data on the cloud at any time.
There are high chances for information to be misused and also deleted either
purposefully or accidentally.

3. Compliance:

Although there are many regulations put-forth related to hosting and data services,
compliance monitoring is very difficult. More effective compliance monitoring
systems such as Federal Information Security Act are very costly for the consumers
to adopt.

4. Sustainability:

This refers to the minimization of the effect of cloud on the environment. Referring
to the servers’ effects on the environment is large. These include availability of
climate dependencies such as natural cooling and renewable power resources.
Countries that has natural environment such as Finland and Switzerland are showing
more interest to implement data centers over the cloud.

5. Abuse:

While giving cloud as service, it is mandatory to be ascertained when the client do

not purchase the service of cloud for any illegal purpose. In the year 2009, a banking
 malware used the Amazon web services and issues software updates and infectious
instructions to PCs which made the system to collapse and loss of data.

6. Resource / Task Allocation

There exists a scenario in cloud where many users are enabled to opt for resources
from available cloud services and hence the allocation of resources to the users as and
when it is requested by the users are an important task [8-9]. In cloud computing,
often many jobs need to be executed at the same time with optimum usage of
resources. Hence allocating resources through efficient task schedulers is a
challenging issue in cloud computing.

1.3 TASK ALLOCATION PROCESS

Cloud computing has paved way for bringing out a dramatic change in the traditional
methods of service deployment by either an organization or as an individual. It offers
various types of services for the users registered in form of web services making the
investment on computing infrastructure minimum. In each of the service being provided,
the users normally submit requests to the CSP over the Internet. The CSPs are in turn
responsible for the allocation of the requested resource from the users [10]. Different
scheduling methods are adopted by the service providers for scheduling the incoming
tasks so as to ensure effective management of the resources. Task schedulers and
optimum resource usage helps the providers to attain maximum revenue through optimal
resource usage. It is common that the performance of a cloud directly depend on the way
in which tasks are allocated and resources are managed. The process of task scheduling is
presented in Figure 2.
 Figure 2 – Task Scheduling Process

Task scheduling is termed as the process in which the incoming requests for resources
arising from the users are organized in a certain way to ensure optimal usage of resources
[11]. Every service deployed in the cloud has a range of users and hence at a given point
of time, a number of requests are likely to be generated from the users. The absence of a
good scheduling mechanism may end up in long waiting period for a resource and also
the short termed tasks will get terminated. A range of constraints have to be met at the
time of scheduling which may include the nature of a task, its size and the effective load
on the resources. Scheduling the tasks is important issue to be addressed in cloud
computing. The core advantage of cloud computing is that it extends proper usage of the
resources. Hence, task schedulers and resource allocators are considered as the two sides
of a coin given which are interdependent.

The entire stack of services available can be used by the consumer ranging from different
hardware to web applications. The services of a cloud normally are assigned on a pay-as-
you-go basis. The end user can opt for more or less services as per the demand of the
applications and a pre agreed terms and conditions. The users are given flexibility to rent
any service as they need and to release once the usage is over. The users are given
complete freedom on the choice of the service and the time period for utilizing the same
as per the need [12]. This freedom of choice given to the end users has created some
overheads particularly when the next waiting user cannot be predicted. Hence,
 scheduling of tasks and resource allocation are important things to be considered in cloud
computing research. How efficiently the resources are used are normally dependent on
different load balancing techniques instead of allocating the resources in a random
manner. Cloud technology is used to solve complex tasks. In order to solve complex
issues, a proper scheduler is recommended. These can help in leverage of resources. A
large community of researchers has been actively involved in the domain of task
scheduling in cloud however, there still exists a large demand in effective task schedulers
to cope up with the growing demand.

1.4 RESEARCH MOTIVATION

The task scheduling process plays a critical role in cloud computing. The users are
expected to pay for the resources that are being requested for solving a particular task.
The need of efficient task schedulers are growing more and more as the nature and
complexity of the resources demanded by the users are also rising exponentially. In order
to achieve more profit by the CSPs, it is essential that the task allocation and resource
utilization are done in an optimal manner. Given that there is rapid raise in the number of
users, the tasks are also need to be scheduled in a proportionate manner. The algorithms
used for the task scheduling process vary along different environments [13]. Task
schedulers are aimed to minimize the makespan by making use of the resources in an
efficient manner. The Dynamic load balancing techniques needs to allot tasks to the
appropriate processors as and when they arrive. The redistribution of tasks has to be
made effectively when there are more loads on the processors. Appropriate scheduling is
to be done based on various parameters which ensure good performance of the cloud.
Task scheduling on the cloud is the most important issue which guarantees the cloud
connectivity ensuring that the needs of the consumers in terms of resource allocation are
met. An effective scheduler is expected to comply all the needs of the users thereby
providing QoS [14]. Task planning is expected to focus those activities which enhance
the framework in an effective way. The major hindrance in this process is the assignment
of processors in this heterogeneous framework. This have drawn decent attention from
many researchers to contribute on effective task management in cloud computing.
 The process of task scheduling in a cloud environment is often considered as a
NP-Complete problem. Efficient methods of task management will help to balance the
load on VMs and thereby to achieve effective usage of resources [15]. Hence, there is a
high demand for effective schedulers which can minimize the loads on the processor to
ensure QoS. Hence, as the process of task scheduling is one of the major challenges in
lieu of resource allocation and has great impact on the performance of cloud computing
and the fact that task scheduling process needs to be improved for ensuring quality and
for reducing the cost of computations is taken as the motivation for this research and the
proposed research aims to resolve it in such a way that it gives mutual benefit to the
consumers as well as the service providers.
In many research scenarios, the problem of optimization is non-linear with multi-
objective solutions. These are difficult to be solved even with the most increased
computational powers of modern computers. It is still being impractical for using brute
force methods. Thus efficient algorithms are always in great demand but the truth is that
such algorithms do not exist in most of the optimization problems [16]. Although there
are large number of optimization methods available, these are normally gradient based
where the issue is attaining local maximum and do not attain global maximum. This also
led to a situation where the final solutions always depend on the initial solutions. Also,
the computational complexity will be very huge making the system expensive.
A large range of research problem in Science and Engineering can be formulated
as optimization problems that are subjected to large and complex constraints that are
typically non-linear. The solutions for these non linear problems will be requiring
complex optimization algorithms. In the recent past, nature inspired algorithms are
largely used to solve these optimization issues owing to its flexibility and effectiveness.
In the recent times, significant progress has been shown by nature inspired algorithms in
solving complex optimization problems. Despite its popularity and effectiveness, these
also suffer from some impracticality [17]. First, these lack a unified mathematical
framework which leads to lack in deeper understanding as how these will converge and
how quick this will happen. Next, most of algorithms of this kind are largely based on
numerical experiments which make it difficult for them to be compared. At last, although
 these are found to be effective in solving small scale problems, there is always an
uncertainty as whether it will solve large scale problems. Hence the motivation of the
research is still more focused as how to effectively make use of the nature inspired
algorithms in the process of solving the task scheduling and resource management in
different cloud environments.
1.5 OBJECTIVES OF RESEARCH

The research was conceived based on the intense preliminary studies and various
feedbacks form the Cloud Service providers and clients on the performance issues. It is
clearly evident that even though the technology claims that it has major advantages and
easy deployments, few hindrances and hesitations were still persistent for the CSPs and
for the companies or organizations to move their businesses into cloud. There was always
a key concern on the complex task allocation process particularly when there is a
growing number of requests for a specific resource and effective handling of resources.
The research spark came out with in an intention as to how technically contribute to the
enhancement of task management and resource management in cloud. Having kept the
intentions of the research clear, the objectives of the research were finalized which are as
follows

1. To study and investigate the various computational and theoretical models

adopted for the effective task scheduling in cloud computing
2. To understand the implications of nature inspired algorithms in the task
management domain of CC and to extract the pitfalls in existing methods
3. To resolve the issue of multi-objective task scheduling problem using the nature
inspired Squirrel Search method for IaaS cloud model
4. To maximize the profit in Task scheduling using the nature inspired Whale
Optimization in Hybrid cloud
5. To implement MAPReduce and Genetic Algorithm(GA) using the Whale
optimization for better task scheduling in cloud
6. To propose a novel cloud infrastructure for embedded system domain taking the
task scheduling as priority.
1.6 PROBLEM ANALYSIS
With the objectives cited above, intense study was carried out to identify the narrowed
down problem to be addressed which are described as below
1. A large number of users share the resources in cloud by submitting the computational
tasks to the cloud systems. Scheduling all these task turns out to be a challenging process
in a given cloud. As the computing facilities are finite, there is a great need for an
efficient resource management technique. One of the common problems in achieving this
is that when these algorithms are designed for avoiding overloading on any given
processor, the runtime tend to increase leading to lot of energy usage.
2. As far as the VMs are considered, the single-target solutions are normally used which has
certain issues. For example, the assignments which have more priority tend to have more
chances of being executed first than the tasks which as less priority. The less prioritized
tasks also have chances to be executed but it happens only after the high priority tasks
being executed. This can make the low priority to starve for an indefinite time when the
high priority tasks are continuously flowing in. This works fine in most of the scenarios,
but reduces the performance when we consider the total tasks requested and the number
of tasks completed. In order to solve this, effective algorithms are needed which can
optimally solve this in both the scenarios. There is also a need for an efficient planning
algorithm in the schedulers to ensure high efficiency. Hence, tasks need to be effectively
assigned through robust algorithms for reducing the cost and time.
3. As and when the total users increase in a cloud system, a proportionate method needs to
be adopted for task scheduling. The choice of task schedulers will be different for
different environments. There is a need for dynamic load balancing methods for
allocating tasks to the computational processors. Appropriate methods need to be
followed for the re-distribution of tasks when there is more loads at a given processor.
Any cloud service provider is intended to get more profit without violating the standards
and the services that are promised to an end user. Scheduling often needs to ensure
various parameters to be satisfied for overall performance. It is seen that there is a large
research gap exists in focusing on the profit maximization in task scheduling
4. Random search oriented schedulers are largely adopted in cloud task scheduling owing to
its robustness. However, these are more sophisticated for implementation and hence
incur more cost and complexity and also could not adopt with the dynamic requests from
the users making it almost impossible to be adopted. Several parameters have to be
considered before using a task scheduler like the time complexity, makespan, load
balancers and tolerance ratio and QoS. Meta-heuristic techniques are currently used in
large scenarios for obtaining a near-optimal solution for large and complex issues. The
various meta-heuristic algorithms, although these proved efficiency and effectiveness,
there is always a good scope for the improvement in the performance aspect for task
scheduling in a cloud based service. Usage of MAP Reduce frameworks along with
Genetic Algorithms were not tried in the earlier methods.
5. The Round Robin algorithm is considered to perform well in case of many CPU
scheduling problems and have achieved great performance through fair allocation
policies which are based on the time frames in each of the processes. The processes that
are running are forced and preempted from the CPU into a queue and hence avoids
starvation in the system by making a long wait time for the processes to be allotted CPU.
If there are N tasks, then every task will be allotted 1/N of the VM’s processing time and
the waiting time will not be more than (n−1)∗QT and the overhead of scheduling will
be of the order of o(1). But, the Cloud environment are subjected to face many changes
while the tasks wait for a longer time in the queue and fixed time slicing method cannot
assure best system performance. The time of wait and the ratio of the context switching
need to be handled cautiously. Hence this is to be taken as an optimization problem and
an enhanced method is needed to overcome this issue.
6. A static algorithm normally does not depend on the present state and hence there should
be a prior knowledge on the system. These static algorithms suits only where the
workload does not change quite often. Hence, these are unfit to be used in cloud where
the workload is dynamic. This is considered as a NP complete problem which cannot be
solved polynomially. Robust optimization methods are needed for the best solutions to
these problems. Hence, major change in the framework of task schedulers is needed for
achieving best QoS.
2. LITERATURE SURVEY
The most recent research papers, journal articles, white papers and books were studied to
get a decent knowledge on the methods utilized for overcoming the issues on task
scheduling and allocation in cloud computing. The search was narrowed down using
specific key words including “Task allocation in cloud”, “Resource optimization in
cloud”, “Nature inspired algorithms for optimal task allocation” and “MAP Reduce
frame work for task allocation in cloud”. The search timeline was also made in such a
way that it covered the publications ranging from first 5 years of the technology being
introduced and the literatures from last five years i.e. from 2014 to 2019. Out of the
hundreds of literature sources, the most relevant were taken and intense study was done
and a brief discussion of them is below.

D. Wu et.al [18] proposed a novel method for task management based on enhanced PSO.
The authors have addressed the present inefficiency of task scheduling in cloud. An
iterative operator for selection of tasks has been presented for solving the optimization
issue. Particle Agent Swarm Optimization is improved in the proposed method. The
method focused on the issue of local optimum. The results have proved that the proposed
method was efficient in terms of the coverage which brought down the cost. The results
also showed significant improvement in the optimization and less time consumption. The
major drawback of this system is the inability to produce better performance when the
number of tasks is more.

Y.Cui et al [19] proposed a workflow for task planning and execution in cloud. The
proposed method was based on Genetic Algorithm (GA). Here, the jobs are given
priority in a top-down strategy. By using this method, the tasks in a given workflow are
split into different stages so as to enable parallel execution. A 2D coding method was
then followed for the task planning. A novel Genetic crossover and mutation was then
implemented where the fitness function makes the separate fitness to get synchronized
with the schedule. The proposed model is then evaluated through the realistic workflow
by simulation. The proposed model achieved good cost cutting measures. The major
disadvantage being the time complexity involved where a task needs to wait for a
particular resource.

Chunlin et al [20] introduced a effective task scheduler for heterogeneous cloud for
ensuring high order of resource utilization. In the proposed method, the task scheduling
was based on the Bi-polar neural network in order to ensure that the given tasks are done
within a pre-defined deadline. The results show that there is a significant reduction in the
response time which improves the overall throughput. The proposed method was also
able to produce more QoS in minimal cost. The disadvantage being that the proposed
method did not gave expected results when deployed in hybrid and public cloud.

H.Yuan et al [21] investigated in the problem as to how maximization of the profit can
be achieved in a private cloud without compromising the QoS and delay tolerance. A
profit maximization algorithm (PMA) was introduced which can schedule all the
incoming tasks in a dynamic manner. The sub issue with each of the iteration of proposed
PMA is then solved using the hybrid meta heuristic method of optimization which are
simulated using the annealing PASO. The results when compared to the existing
optimization methods proved that the proposed method is more scalable and robust. The
overall throughput was also seen increasing when excessive simulation is done. The
major hindrance with the proposed method is that it was not able to handle the service
delay bounds.

Kairong et al [22] recommended a technique that used adaptive GA technique for the
optimum scheduling of tasks over the cloud. The proposed method used incremental
Genetic Algorithm that has adaptive measures for the cross over value and the mutation.
These rates have been changed as per the generations among individuals. The tasks were
generated on a random manner for the simulation of different scales in a given cloud
environment. The Amazon EC2 systems are implemented with different VMs of varying
computational capacities on the CloudSim. The experiments prove that the proposed
systems are highly capable of producing optimum solutions with high makespans with
less time complexity. Although efficient, the system did not consider the factors such as
energy dissipation and resource usage.
Mukherji et al [23] introduced a task scheduler based on multi-class decision makers for
the cloud environment. The system makes use of NLP techniques in cloud environment
for the minimization of cost by implementing a queuing system. The proposed method
assumes that a batch contains a constant number of batches with tasks in order to
calculate the time of service through Gantt charts. The next step was made to reduce the
number of clusters in a queue by making use of a MMCK model that reduces the average
time for waiting. The experimental results from the simulation shows better results when
compared to that of the other state of the art methods. The scope for improvement was
there as the proposed system produced an average of 15% increase in the performance.
The dead-line scheduling was not taken into consideration by this method.

Agarwal et al [24] proposed a novel method for task scheduling. The Generalized
Priority Algorithm (GPA) was proposed and was experimented versus the RR and the
FCFS Schemes. The number of VMs was dynamically changed and the results are
monitored using CloudSim. The results proved better efficiency than RR and First Come
First Serve (FCFS) methods. In 2015, Lakra et al [25] introduced the Multi task
objective scheduler in-order to correctly map the tasks with the VMs. Unlike the single
criteria methods which take the execution time alone, the proposed method takes
additional performance parameter such as cost, user bandwidth, etc. The results obtained
through the CloudSim tool proves that the proposed method gives more throughput
without being deviated from the SLA. In 2016, Hammad et al [26] proposed the TS-GA
a type of genetic algorithm for solving the scheduling issue of the cloud. The method
aims to minimize the execution time and to maximize the used resources. The results
prove to be efficient in terms of speedup and resource sharing than traditional GA and
RR algorithms. In 2017, Madni et al [27] made a comparative analysis of the popular
heuristic methods based on top 6 parameters. The Min-Min method was found to be
effective than all other methods. The heuristic property was then taken as a base and
many other child algorithms were presented.

R. N. Calheiros et al [28] designed a new hybrid technique for the problem of real-time
scheduling of loads in computing clusters. The proposed method makes the process of
scheduling to get distributed between couple of components namely the admission
controller and the dispatcher. The proposed method also partitions all the tasks using the
DLT method and used the All Node Assignment (ANA) kind of policy clubbed with the
Minimal Node Assignment which effectively reduced the overhead of scheduling. The
results claim to have more efficiency in terms of average waiting time and average
turnaround time. The system could not however handle the indefinite waiting of low
priority tasks.
Wei Lin et al [29] proposed BATS (Bandwidth aware task scheduler) for the task
scheduling across multiple VMs. The algorithm was designed to consider the bandwidth
of the network as an additional resource along the memory and CPU. The proposed
method was also found to optimize the resources. The results were significant in terms of
optimization parameters such as the precision and recall, where as the total time for
completing all the tasks put together i.e. the Makespan was not, improved which is the
major requirement in any task scheduler.
Hong Xu et al [30] proposed the extended bee method which uses the divisible load
balancing technique which was inspired by the nature based foraging tendency of bees. A
novel agent model is proposed for the reduction in delay of network and to significantly
increase the throughput. The proposed model was developed based on the Transportation
problem and was intended to come up with a cloud environment which is fault tolerant
and thereby making the profit high and by making the cost of execution minimum. The
average waiting and turnaround time was also found to be minimum. The proposed
model however was still in theoretical stage and was not tested in real time.
From the literature study, understanding the research gap and enhancements that are
likely possible in the task allocation problem of cloud computing was done. The
methodology adopted, results and drawbacks were analyzed to narrow down the research
topic. It is understood that the task and resource allocation in cloud computing is still a
large concern and there are still many contributions that can be made in this domain.
 3. PROPOSED METHODOLOGIES
3.1 Nature Inspired Chaotic Squirrel Search Algorithm (CSSA) for multi objective task
scheduling in an IaaS cloud computing atmosphere

Taking the first problem statement into account, the nature inspired Chaotic Squirrel Search
Algorithm (CSSA) was proposed for resolving the issue of optimization in a multi-tasking
scheduling for an IaaS cloud. The proposed method creates more job plans unlike the other
methods that are costly to implement. A messy optimization method was followed to attain
global maximum. The method is compliment to SSA where it fails to continue the global
maximization attained. The proposed method was intended to solve the following issues

1. A variety of population is maintained here to solve the issue where people tend to get
dispersed in the given search space during the initial phase of optimization.
2. The convergence velocity is improved by identifying the difference among the people in
a given population by hunting down the elites.
3. The local maximum is avoided by complimenting the demographic differences inside the
given search space.
4. Overall, the proposed CSSA is aimed to bring out significant increase in the performance
of SSA by its compliment, the CSSA.
The pseudo code of the proposed method is shown in Figure 3.

ALGORITHM 1 Pseudo code of CSSA

Input: Population
Output: Pbest (optimized best fitness value)
Compute:
1. For q = 1 to Q (Q is used to execute total generation)
2. Fitness value of the population is estimated
3. By using the jumping search method update the population
4. If q = = Q/n (n denotes total sub stages)
5. Linear regression equation is computed based on three regions
[0.1Q, 0.15Q], [0.15Q, 0.2Q] and [0.1Q, 0.2Q]
 6. If two or more slopes computed positive then
7. Optimize using progressive search method
8. else
9. Optimize using jumping search method
10. End
11. End
12. End

Figure 3 – Pseudo code of proposed CSSA

Progressive Search

The robustness of the proposed CSSA is increased using the progressive search which
works better than the jumping search. Unlike the jumping search, distinct details are
q
presented in the progressive search method proposed. In a given population, Pr ≥P min
then, during the winter season, the strategy of searching remains similar to that of the
jumping method that is dependent on the present implementation of the search criteria.
Where as if the people face threatening, the dimensions are then changed in a random
manner and subsequently, the speed of convergence gets slowed down however it
ensures more data preservation. Alternatively, when the people are safe during the
summer, the search is performed in a similar manner as that of the jumping search but the
Le ' vy( x) flies are replaced by the gliding steps. If certain people are under risk in a given
population, mutation is carried out on only one aspect which ensures that there is a
population variation constantly maintained. The pseudocode of the proposed progressive
search is presented in Figure 4.
ALGORITHM 2 Progressive Search Method

Input: Population
Output: Pnew
Compute:
q
1. If Pr ≥P min

2. If
b≥J xy
 3. Season = winter
4. else
5. Season = summer
6. End
7. If Season = = winter
8. For H1= 1 to Psize (Psize is the total number of squirrels)

9. GP q+1 q q q
k =GP k + w a ×Y d ×(G t −GP k )

10. End

11. If (
b< J xy )

12. Change between

GP L andGP M and one dimension random selection

13. End
14. End
15. End
16. If Season = = summer
17. For H1= 1 to Psize

18. If (
b≥J xy )

19. GP q+1 q q q
k =G t + Le ' vy ( x )×(G t −GP k )

20. End

21. If (
b< J xy )

22. Change between UB and LB and one dimension random selection

23. End
24. End

Figure 4 – Pseudo Code of Progressive Search

3.2 Profit Maximization based Task Scheduling in Hybrid Clouds using Whale
Optimization Technique
Taking into consideration of the problem statement, a novel metaheuristic method is
proposed which is inspired by the nature’s bubble-net hunting mechanism of the whales,
commonly known as the whale optimization. The WOA is applied for solving the task
scheduling. The proposed architecture is presented in Figure 5. The proposed framework
consists of two frames namely for the private and for the public cloud. In case of private
cloud, numerous computers are being pooled for providing virtual resources such as
memory, storage and network. The component allocated for the monitor oversees the virtual
pool of resource and then reports back with the updated information to the scheduler. The
arrival of a particular task from a certain user is then put in queue on FCFS basis. The
complete information on the queue is sent then and there to the scheduler. To add with, the
predictor helps in executing the typical prediction techniques for the information to be
predicted on the public and private cloud with the help of the historical data. The important
contribution of the proposed technique lies in the consideration of scheduling component
which executes the temporal schedule of tasks for maximum profit in a private cloud and
thereby to guarantee the service delay bounded with all the other tasks. The proposed WOA
is then compared to that of the other competitive algorithms such as the Artificial Bee
Colony (ABC) and the Genetic Algorithm.
 User’s task

FCFS
queuing
Public cloud 1
Task predictor
Resource monitor

Task Scheduler

Public cloud 2
CPU Memory Network Storage

Private cloud Public cloud N

Figure 5 – Proposed Architecture

The WOA is made used for the selection of optimum Final score based on the values of
scores that are obtained from optimal tasks. The proposed WAO acts on three important
strategies namely the encircling prey, bubble net attacking and the search strategy of
prey. The flow chart of the proposed method is presented in Figure 6.
 Start

Initialize the number of task, virtual machine & physical machine

Evaluate fitness function (i.e.) Maximum profit

Using eqn (24) update

No
current search agent If R<0.5
position
Yes
Using eqn (23) update No Yes Using eqn (21) update
current search agent If A<1 current search agent
position position

If f < max Yes

T=T+1
iteration

Save the maximum profit value

Stop

Figure 6 – Whale Optimization

It is seen that in each of the iteration, the agents meant for search updates the respective
positions in consideration to either of a random search agent or from the best ever
solution which is obtained so far. The parameter a is then decremented from a value 2 to
0 for providing exploration and then exploitation. The searching agent is then chosen

when|V|>1 , and the better solution are extracted when the |V|<1 for the updated position
of searching agents. Depending on the R value, the WOA will be able to switch in-
between in spiral or in a circular motion. At last, the algorithm is terminated when
reaches a satisfaction criteria. At last, the value of score is obtained from WAO based on
the score values obtained on the tasks that has been selected. Hence, maximum profit is
ensured by the selection of optimum tasks.
3.3 Enhanced Round Robin (ERR) algorithm for Effective and Efficient Task
Scheduling in cloud environment
The problem statement 4 is resolved by proposing an enhanced RR which solves the
bottleneck of the traditional RR. The traditional RR is modified for enabling it to make
dynamic decisions by making suitable adjustments to the time quantum based on a given
situation which can give positive signs in terms of response time. Here, ERR is presented
and a comparison is made between the RR and ERR on same experimental conditions.
The task scheduling concept is based on the dynamic allocation of time quantum. Here,
the time slices are calculated in a dynamic manner and hence are subjected to vary based
on the task considered. The time quantum as shown in figure will be different for
different tasks. The pseudo code of the proposed method is shown below
PSEUDOCODE

IP: Cloud (tasks), VM

OP: map each task to VM

Steps:

1 NO_CL : List(cloudlet).Sizeof() No_VM: VM.Sizeof()

2 I= 0
3 FOR J=0 ; No(CL) do
4 CL.list.getlist ();
5 Index = Index +1 ; No_VM VM.get(indices)
Stage(transfer_time(CL, VM,Index) Out.trans_time(Cl , V)
6 Ex. Exec_time(Cl.V)
7 IF Cl.AT+Stage.in+exec+sta<=Cl.Dl THEN Send_job(Cl.V)
8 Update (V) Else Drop (CL) Failed(tasks)
9 Stop

Figure 7– Pseudo code of Proposed ERR

The ERR proposed above is a minor variant of the traditional RR as it considers the time
slices in a dynamic manner for each task that enters into the queue. The time slices are
assigned based on the Mean Burst time of all the waiting tasks in the queue. Couple of
memory registers namely the SREG and AREG is used for achieving this. The former
stores the total burst time required and the latter denotes the average burst time. In the
beginning, the first task is allotted to a VM and it takes all the time. Then, the scheduler
does the job of calculating the time slice required for the entire incoming task thereafter.
Every task when allotted a VM will be running for a certain time quantum and when it
lapses, it is either moved back to the queue and the completed task will be removed out
of the queue. The scheduler hence adjusts the register values by subtracting the time
quantum from the SREG and by adding the same to the other register AREG.

3.4 An infrastructure for embedded systems using Task scheduling

The final research problem is addressed by proposing a preemptive scheduling with

honey bee foraging (PSHBF) algorithm to optimize task scheduling. The resultant
optimum solution has a set of Virtual machines for which the given tasks are
preemptively scheduled. The proposed technique takes into consideration of the task’s
priority and the expected time of completion of the remaining tasks during the process of
preemption. The preemption of tasks is performed to minimize the time taken for
response and for execution of tasks that has different priorities. The proposed method
follows a foraging which is carried out by the honey bee for the selection of best solution
among the resources that are available and hence achieves a load balance in the cloud
data center. Presently, the core attention is made on performing an intra-host kind of load
balancing on the VMs inside a single physical host. From the deeper study of the
foraging of bees, the mapping in-between the load balancing of the system proposed and
the foraging of honey bee as presented in Table 1.

Food source : Low loaded VM group

Nectar : VMs in LVM group
Employed bees : Tasks running in each VM
Onlooker bees : Tasks in queue/new incoming task
Scout bees Tasks removed from a VM
:
belonging to OVM group
Information shared through Load and capacity on each VM and
waggle dance : the total number of tasks and the
priority of tasks in each VM
Table 1. Mapping between scheduling of PSHBF with Honey Bee behavior
Preemptive Scheduling

The Figure 8 presents the process of working of the proposed PSHBF. As per the
proposed method, the balancing of load is done by the removal of tasks from the VMs
that are overloaded and by scheduling them under the VMs that are loaded. Hence, task
scheduling is carried out in a preemptive manner through the consideration of the priority
in tasks during the preemptive process. Three classes on the priority namely the high,
medium and the low are taken for consideration. While removing a particular task from
the VM that is overloaded, both the priority and the time of completion are taken into
account for minimizing the makespan and response time for a given task. The following
possibilities are taken into consideration in preemptive scheduling.
1. To consider the tasks with high priority for the preemption (or)
2. To exempt the tasks with high priority for preemption
The core reason to identify a second chance is when a task with more priority which is
currently being executed are preempted, it creates a large impact on the satisfaction of
the user and the time for response can end up in a non-achievement of Quality in service
from the service provider. As an outcome, the risk of SLA being violated will be more.
The Figure 8 presents the preemptive task scheduling. The preemption happens only if
the priority of task that is removed is more than the current task which is running in the
loaded VM where as the remaining expected time of completion is large when compared
to the burst time of the task being removed. Once all the conditions are satisfied, the
process of preemption happens and the task that is removed and the tasks that are
interrupted are restored from the pausing point [23]. The checkpoints are used to oversee
the status of the tasks. When a given task is allotted to a VM, the information is updated
in a way that the total tasks that are assigned, their priorities and the updated VMs
(OVM, UVM, BVM). The VM will be placed in the BVM till it reaches a threshold. The
load balancing will be successful once every VM is moved to the BVM.
 Preemptive task scheduling

Is incoming task’s priority > = priority of running task?the incoming task in to target VM
Queue

Is expected remaining execution time of incoming task < running task’s remaining execution time?
Execute the running task

Save the state of running task

Execute the incoming task

Figure 8 – Proposed Preemptive scheduling

The proposed technique executes a preemptive scheduling by initiating a foraging

character of honey bee and thereby performing load balancing on VMs. The QoS is also
maintained by considering the priority and the expected completion time of the set of
tasks taken under a given moment.
4 RESULTS AND DISCUSSIONS
4.1 RESULTS OF THE PROPOSED CSSA
The experiments were performed in CloudSim toolkit and the table 5 shows the result
obtained over the methodology proposed in 3.1.1.

CSSA SSA PSO-BAT Hybrid GA-PSO

Energy 5.1 7.4 6.7 9.33

Time 930 1334 1208 1682

Cost 468 539 577 657

Resource 0.86 0.6 0.82 0.59

utilization

SLA Violation 13 40 17 40

Table 2 – Evaluation results of proposed CSSA

The above table 2 presents the final comparative analysis of the proposed CSSA with that
of its close competitive methods namely the SSA , PSO-BAT and the hybrid GA-PSO. It
is evident that the proposed CSSA is found to have minimal energy value of 5.1 whereas
the other have a significant difference from the proposed method. The time taken for
execution of the CSSA is found to be 930Ms which is very lesser than the other methods.
The computational cost obtained is 468 for the CSSA whereas the other methods have a
cost of 539 , 577 and 657 which proves the robustness of the proposed method. It is also
known from the table that the CSSA has the lowest utilization of resource with 0.17 than
the SSA(0.38) , PSO-BAT(0.24) and GA-PSO(0.3). The propsoed CSSA also have a
minimal violation rate with 13 which is 16% better than the other methods on an average.
Hence, it is concluded that the CSSA has minimum energy usage, cost and resource
utilization which ensured optimal planning of tasks in an VM.
4.2 RESULTS OF THE PROPOSED WHALE OPTIMIZATION METHOD
The outcomes of the experiments from the proposed WOA are presented below. The
experiments were conducted using the JDK tool kit 1.6 taking the CloudSim as a tool for
simulation. The synthesized data are used as the input for effective task scheduling. The
evaluation metrics considered are the efficiency, profit, time and memory. The overall
results are presented in Figure 9.

Figure 9– Comparative results of the Proposed WOA

It is understood from the figure that the mean efficiency obtained using the proposed
WOA is 78.73% which is significantly larger than the ABC method (78.03%) and
Genetic Algorithm (77.30%). The average profit was also found to be more (71.95%)
than the other methods. The average time taken to schedule 10 tasks was found to be
12364 Ms in case of proposed GA-WAO whereas the same is 12411Ms in case of ABC
and 1273Ms in case of GA. The efficiency was also proved by a lowest processing time
of 14449.5 ms in case of proposed method whereas it is much more in case of the other
comparative methods. The proposed method also outperforms the ABC (151749.3 Mb)
and GA (15648.3 Mb) by producing a lesser Average waiting time of 137158.3 Mb.
 Figure 10 - Comparative analysis of performance metrics of WOA
The proposed method achieves a large profit of 2426.2$ which is significant than the
other methods such as GA (2212.8$) and WOA (2426.2$) in case of private when the
number of iterations is fixed as 100. From the figure, it is proven that the proposed
method is more efficient in terms of maximum time taken for a given task to be
completed (160.9 sec) with a profit of 1000$. The mean value of Memory and CPU used
are high in the proposed WOA method (52.6435%) where as the comparative methods
like GA and ABC has only lesser usage of CPU which causes delay in task processing.
4.3 RESULTS OF THE PROPOSED MAP REDUCE AND GA BASED WHALE
OPTIMIZATION
The proposed system is subjected to intensive experiment in the CloudSim tool for the
evaluation. The system is compared with other state of the art methods in the literature on
various parameters including mean wait time, mean turnaround time, latency and average
throughput. The analysis of the same is presented below. The proposed GA-WOA is
compared with other competitive methods such as the WOA, Squirrel Search
Optimization (SSO), Elephant Herd Optimization (EHO) and Grey Wolf Optimization
(GWO).
 Figure 11 – Comparative analysis of performance of proposed GA-WOA

Figure 11 presents the comparative results obtained over experiments. The number of
tasks are kept as 100 and it is seen that the proposed GA-WOA takes only 220ms of time
where as the GWO had a mean turnaround time of 498ms. Similarly, when the number of
task is increased to 200, the average turnaround time was 371Ms which is relatively
lesser than the other methods such as WOA (411 Ms), SSO (482 Ms), EHO (562 Ms),
and GWO (683 Ms). Latency is taken as one of the other parameter metric. It is defined
as the time taken for completing a task taken. The experiments produced a best latency
for the number of tasks ranging from 100-500 for the proposed GA-WOA (872 Ms,
1885Ms, 2706Ms, 3892Ms and 5404 Ms) which was found very lesser than the other
methods taken for comparison. The cost involved in computation of other methods was
1053 on an average but when MAPREDUCE functions are used, the same is reduced to
876 which proved the efficiency of MAPREDUCE being used. Also for a given 500
tasks, the average cost of communication was found to be 867 and 837 with and without
MAPREDUCE being used. Thus implementation of MAPREDUCE in the proposed GA-
WOA produces more efficiency.

4.4 RESULTS OF THE PROPOSED ENHANCED ROUND ROBIN ALGORITHM

The experiments were conducted in CloudSim environment which provides the all
functionalities as that of a real cloud environment. Specific scheduling mechanisms can
also be simulated with customizable options. The efficiency and the performance can
also be tested here. The proposed MMR is tested in the toolkit with different values for
VM and compared with that of traditional RR in the same VM environment. The no of
VMs were altered exponentially and simultaneously the cloudlets (tasks) were also
increased. The results are discussed below. Table 1 show that the AWT of ERR is found
to be much lesser than that of the RR. The size of the time quantum is increased up to 5
times the initial value. Table 2 shows the comparative analysis of the performance of
other scheduling methods with that of the ERR and the proposed method is found to be
better in terms of Execution time per 100 cloudlets considered and Residual energy. The
same is depicted in Figure 12.

No of tasks AWT in RR AWT in Relative Error

ERR
5 5 4.21 0.16
6 7 6.11 0.12
7 9 7.62 0.12
8 12 9.08 0.04
9 16 12.18 0.09
10 21 17.16 0.12
Table 3- Comparison of AWT

Methods Evaluation Measures

Execution Time Residual Energy
ACO 16.12 18.15
 GA 15.26 17.62
MPA 14.85 19.28
FCFS 17.12 13.95
PSO 12.15 16.62
MRR 08.18 10.18

Table 4- Comparison of Performance metrics

Figure 12 – Comparison of AWT and performance of ERR

4.5 RESULTS OF THE PROPOSED PREEMPTIVE SCHEDULING WITH

HONEY BEE FORAGING (PSHBF) ALGORITHM

For the experimental analysis, a single data center is taken with 20 numbers of hosts
where each of them runs 4 VMs respectively. Around 300 numbers of cloudlets were
created for scheduling and for execution. Each of the node consisted of single CPU with
a minimal capacity of 1000 MIPS. The evaluation metrics considered for the
performance evaluation are makespan, response time and the time for execution with the
total number of mitigated tasks. The results are observed by changing the tasks in an
incremental manner i.e. from 50 to 300. Here, the heterogeneity is maintained by
changing the length of every task and the heterogeneity among the resources are
achieved by changing the capacity of the processor on every VM. The experimental
results are presented in Figure 13.

900
Average Execution

Average Response time

800 400
HBB-LB (non-preemptive)
HBB-LB 350
700
Time

(non-preemp-
600 tive)
300
PSHBF (considering high
500
250 priority tasks)
PSHBF (ex- 200
400 empting high
150 PSHBF (exempting high
300 priority tasks)
100 priority tasks)
200
50
100
0
0 50 100 150 200 250 300

50 100 150 200 250 300 No. of Tasks

No. of Tasks Figure 4. No. of tasks vs.Average Response time

Average Makespan

No. of Tasks Migrated

HBB-LB (non-preemp-
1000 HBB-LB (non-preemp- 140 tive)
800 tive) 120
PSHBF (considering 100 PSHBF (considering high
600
high priority tasks) 80 priority tasks)
400
PSHBF (exempting high 60 PSHBF (exempting high
200
priority tasks) 40 priority tasks)
0 20
0
No. of Tasks
Figure 5. No. of Tasks vs Average Makespan
No. of Tasks
Figure 6. No. of tasks vs. No. of task migrated

Figure 13 – Comparison of results out of proposed Preemptive task scheduler

It is seen from the figure 13 that the proposed PSHBF minimizes the time for execution
to 13.5% and in case 2 it is reduced to 14.7% when compared to that of the competitive
HBB-LB. The time taken for execution is increased as the number of tasks is increased
owing to the constant VMs. As VMs are considered as the exact resources which execute
the tasks, the increase in total count of the VMs can still more minimize the execution
time. When the response time is considered, the proposed method when compared to
HBB-LB gave 18.3% improvement in case 1 and 21.9% in case 2. As the response time
is considered as one of the metric of user satisfaction, it is concluded that the proposed
PSHBF gives more satisfaction than the HBB-LB.
The second trial of experiment were performed by taking only the high prioritized tasks
where it was seen that the number of task migrations were less in the proposed method
than HBB-LB. It is seen that for 50 numbers of tasks, the total tasks that are migrated in
case of HBB-LB are 10 and it was minimized to 5 and 4 in case 1 and 2 when PSHBF is
considered proving the overall efficiency. Also, when the case 1 and 2 of PSHBF is
taken, the case 2 presents notable improvement when compared to case1 when there are
less number of tasks. This is because that in case 2, the tasks which have high priority are
not rescheduled where as only the ones with low and medium level of priority are
removed and re-scheduled. When the total number of tasks are increased, there was seen
only a minor deviation of the results obtained through case 1 and 2. Hence, taking the
task priority as an important factor of performance, it is proved that the PSHBF is more
efficient in case 2 than in case 1.
5. CONCLUSION

Cloud computing is turning out to be a significant technology in the parallel and

distributive computing paradigm and have attracted a large research community. A large
volume of users are attracted towards the cloud computing owing to its computational
capabilities at low cost and less infrastructure overheads. Various services are offered in
cloud which includes storage, infrastructure, platforms and software in the medium of
Internet. The important models of cloud computing are IaaS, PaaS and SaaS respectively.

Huge number of users shares the resources over cloud by submitting the task to the cloud
system. It is a huge challenge for the CSPs to schedule these tasks in the cloud
computing environment. Cloud Computing has come up as an emergent technology from
distributed, parallel and grid computing. As the clouds are designed for providing
services for users from outside, the service providers needs to be compensated for the
services being provided. As the resources are finite, there should be an effective
mechanism followed for their optimal usage. Effective management of resource will
ensure the minimization in the number of VMs required and which in turn can reduce the
carbon footprint which could be a great energy saver. Complex schedulers can result in
minimized overload but also can end up in more waiting time for a particular resource.
Taking the importance of the task scheduling and the pitfalls in the existing methods, the
research is presumed to bring out a technical contribution in the task scheduling by
making use of the nature inspired algorithms.

In the first phase of the research, a chaotic squirrel algorithm was proposed as a
compliment to SSA as the latter fails to expand the search area adequately in time of
winter and also the searching time during summer is also uncertain for the integration
purpose. The proposed CSSA used leaping and progressive search method. The proposed
CSSA when compared to SSA was made to focus more on the improvement in winter
and also the specialty in summer which maintain an excellent equilibrium among the
growth and also increases the flow velocity with high precision. The proposed method
was also able to find an optimum alternate much easier than SSA. The convergence is
ensured by maintaining a high order of population diversity. The experimental results
also prove that the proposed optimization method can identify a greater compromising
solution as opposed in the latest schedulers.

In the second phase, the problem of solving the task scheduling in cloud is taken and
WOA was proposed for the same. The WOA used the principles of nature inspired
evolution for reducing the redundancy in computations. The profit maximization in a
private cloud was taken as the core objective in this phase. The same is achieved by
using a FCFS queue in a task scheduler. The experiments show that the proposed method
ensured not only the maximization of the profit, but also was able to improve the overall
efficiency by reducing the processing time when compared to that of other competitive
methods such as the ABC and GA methods. The proposed WOA method was found to be
more effective in task scheduling when compared with that of the other methods.

The third phase of the research was conceived by taking the issue with task schedulers
that the attention is provided only for the task execution and the band width of features
for resource allocation whereas the time for sending the tasks that are important
performance parameter are ignored, this phase of research proposed an effective
approach using the MAPREDUCE and GA-WOA for scheduling tasks in an optimal
manner in cloud. The proposed methodology consists of '4' stages, namely, feature
reduction, feature selection, task separation, and task scheduling. Here, the performances
of the proposed GA-WOA algorithm are compared with the existent WOA, SSO, EHO,
and GWO algorithms. The performance is compared based on some quality metrics,
namely, average turnaround time, average waiting time, processing time, latency, and
throughput metrics. Experimentation was executed on CloudSim, which is used for
modeling the different task-scheduling algorithms. The average time for turnaround is
reduced overall. Also, the proposed methodology takes only 5897 Ms for processing a
task with a minimal latency computed to 872Ms. From the experimental results, it can be
concluded that the proposed method gives high performance than the other methods used
in the cloud for task scheduling with existing resources for minimizing the time taken for
processing a given task.

In the next phase, as the scheduler is expected to find in an optimal way in which the
task-VM mapping could be done in a best way such that the time involved in allocation
and response are minimum with minimal make span. In order to achieve this, an
enhanced RR was proposed to solve the bottleneck of the traditional RR. The traditional
RR is modified for enabling it to make dynamic decisions by making suitable
adjustments to the time quantum based on a given situation which can give positive signs
in terms of response time. The ERR proposed above is a minor variant of the traditional
RR as it considers the time slices in a dynamic manner for each task that enters into the
queue. The results show that the proposed method produces less Average waiting time
for tasks and also found to be more effective than the other scheduling algorithms in
terms of residual energy and execution time.

In the final phase of the research, the preemptive scheduling based on honey bee foraging
(PSHBF) was proposed for the optimized scheduling of tasks. The preemption of tasks is
performed for the reduction of response time and the time taken for execution of the
tasks belonging to multiple levels of priorities. The experimental results were found to be
significant when compared to that of the existing HBB-LB method. The results have
proven that the proposed PSHBF can act as an alternative to the existing task schedulers
for performing load balancing and also to improve the overall Quality of Service (QoS).

Thus it is concluded that task allocation and management plays a vital role in
determining the performance of the cloud. Although many algorithms were proposed in
the literature, Nature inspired algorithms were found to produce better results. Even as
such is the case, the pitfalls of those are identified and are addressed in the current
research. The objective of using different nature inspired algorithms for achieving better
performance in task scheduling for various cloud architectures was achieved. The
experimental evaluation stands as a proof for the performance and efficiency of the
proposed methods.

6. FUTURE SCOPE

The focus was made on the issue of task management in cloud in this research. Although
the experiments have produced significant results, different nature inspired algorithms
were used in-order to achieve the best output. In future, the focus will be made as to
bring out a common universal standard for the task allocation across the different cloud
platforms. Also, the algorithms proposed will be subjected to parameter fine tuning for
making it best fit independent of the number and complexity of tasks and resource
requests from the clients. A standard operating protocol for task management is planned
to be proposed in the future which matches with the global standards.

REFERENCES

1. Y. Xiong, S. Huang, M. Wu, J. She and K. Jiang, "A Johnson's-Rule-Based

Genetic Algorithm for Two-Stage-Task Scheduling Problem in Data-Centers of
Cloud Computing," in IEEE Transactions on Cloud Computing, vol. 7, no. 3, pp.
597-610, 1 July-Sept. 2019
2. Moon Y., Yu H., Gil J., Lim J., "A slave ants based ant colony optimization
algorithm for task scheduling in cloud computing environments,” in Human-
centric Computing and Information Sciences 7: 28 (2017)
3. Fei Luo, Ye Yuan, Weichao Ding, and Haifeng Lu. , “An Improved Particle
Swarm Optimization Algorithm Based on Adaptive Weight for Task Scheduling
in Cloud Computing,” in Proceedings of the 2nd International Conference on
Computer Science and Application Engineering (CSAE '18). Association for
Computing Machinery, New York, NY, USA, Article 142, 1–5
4. B. A. Al-Maytami, P. Fan, A. Hussain, T. Baker and P. Liatsis, "A Task
Scheduling Algorithm With Improved Makespan Based on Prediction of Tasks
Computation Time algorithm for Cloud Computing," in IEEE Access, vol. 7, pp.
160916-160926, 2019, doi: 10.1109/ACCESS.2019.2948704
5. Seth S., Singh N., “Dynamic heterogeneous shortest job first (DHSJF): a task
scheduling approach for heterogeneous cloud computing systems,” International
Journal of Information Technology 11, 653–657 (2019)
6. Dabiah Ahmed Alboaneen, Huaglory Tianfield, and Yan Zhang. 2017.,
“Glowworm Swarm Optimisation Based Task Scheduling for Cloud Computing”
in Proceedings of the Second International Conference on Internet of Things, Data
and Cloud Computing (ICC '17). Association for Computing Machinery, New
York, NY, USA, Article 152, 1–7
7. K. Wu, P. Lu and Z. Zhu, "Distributed Online Scheduling and Routing of
Multicast-Oriented Tasks for Profit-Driven Cloud Computing," in IEEE
Communications Letters, vol. 20, no. 4, pp. 684-687, April 2016, doi:
10.1109/LCOMM.2016.2526001
8. Ramegowda A., Agarkhed J. & Patil S.R., “Adaptive task scheduling method in
multi-tenant cloud computing,” in International Journal of Information
Technology (2019)
9. Huynh Thi Thanh Binh, Tran The Anh, Do Bao Son, Pham Anh Duc, and Binh
Minh Nguyen. 2018, “An Evolutionary Algorithm for Solving Task Scheduling
Problem in Cloud-Fog Computing Environment,” in Proceedings of the Ninth
International Symposium on Information and Communication Technology (SoICT
2018). Association for Computing Machinery, New York, NY, USA, 397–404
10. P. Zhang and M. Zhou, "Dynamic Cloud Task Scheduling Based on a Two-Stage
Strategy," in IEEE Transactions on Automation Science and Engineering, vol. 15,
no. 2, pp. 772-783, April 2018
11. Sreenu K., Sreelatha M. , “W-Scheduler: whale optimization for task scheduling in
cloud computing,” Cluster Computing 22, 1087–1098 (2019)
12. Xing Liu, Panwen Liu, Hongjing Li, Zheng Li, Chengming Zou, Haiying Zhou,
Xin Yan, and Ruoshi Xia, “Energy-aware Task Scheduling Strategies with QoS
Constraint for Green Computing in Cloud Data Centers,” in Proceedings of the
2018 Conference on Research in Adaptive and Convergent Systems (RACS '18).
Association for Computing Machinery, New York, NY, USA, 260–267
13. S. Pang, W. Li, H. He, Z. Shan and X. Wang, "An EDA-GA Hybrid Algorithm for
Multi-Objective Task Scheduling in Cloud Computing," in IEEE Access, vol. 7,
pp. 146379-146389, 2019, doi: 10.1109/ACCESS.2019.2946216
14. Valarmathi R., Sheela T., “Ranging and tuning based particle swarm optimization
with bat algorithm for task scheduling in cloud computing,” Cluster Computing
22(3), 11975–11988 (2019)
15. Jiahui Jin, Junzhou Luo, Aibo Song, Fang Dong, and Runqun Xiong. 2011, “BAR:
An Efficient Data Locality Driven Task Scheduling Algorithm for Cloud
Computing,” in Proceedings of the 2011 11th IEEE/ACM International
Symposium on Cluster, Cloud and Grid Computing (CCGRID '11). IEEE
Computer Society, USA, 295–304
16. H. Zhang, J. Shi, B. Deng, G. Jia, G. Han and L. Shu, "MCTE: Minimizes Task
Completion Time and Execution Cost to Optimize Scheduling Performance for
Smart Grid Cloud," in IEEE Access, vol. 7, pp. 134793-134803, 2019, doi:
10.1109/ACCESS.2019.2942067
17. Pradeep K., Prem Jacob T. , “A Hybrid Approach for Task Scheduling Using the
Cuckoo and Harmony Search in Cloud Computing Environment”, Wireless
Personal Communications 101(6), 2287–2311 (2018)
18. D. Wu, "Cloud Computing Task Scheduling Policy Based on Improved Particle
Swarm Optimization", 2018 International Conference on Virtual Reality and
Intelligent Systems (ICVRIS), 2018.
19. Y. Cui and Z. Xiaoqing, "Workflow tasks scheduling optimization based on
genetic algorithm in clouds", 2018 IEEE 3rd International Conference on Cloud
Computing and Big Data Analysis (ICCCBDA), 2018.
20. Chunlin L., & LaYuan L. (2017), “Optimal scheduling across public and private
clouds in complex hybrid cloud environment,” Information Systems Frontiers
19(1), 1-12
21. H. Yuan, J. Bi and M. Zhou, "Multi-Queue Scheduling of Heterogeneous Tasks
with Bounded Response Time in Hybrid Green IaaS Clouds," in IEEE
Transactions on Industrial Informatics, vol. 15, no. 10, pp. 5404-5412, Oct. 2019
22. Kairong Duan, Simon Fong, Shirley WI Siu, Wei Song, and Steven Sheng-Uei
Guan, “Adaptive Incremental Genetic Algorithm for Task Scheduling in Cloud
Environments”, Symmetry, vol. 10, no. 5, pp. 168, 2018.
23. Proshikshya Mukherjee, Prasant Kumar Pattnaik, Tanmaya Swain, and Amlan
Datta, “Task scheduling algorithm based on multi criteria decision making method
for cloud computing environment: TSABMCDMCCE”, Open Computer Science,
vol. 9, no. 1, pp. 279-291, 2019
24. Agarwal A., Jain S., "Efficient Optimal Algorithm of Task Scheduling in Cloud
Computing Environment", International Journal of Computer Trends and
Technology (IJCTT) V9(7) :344-349, 2014.
25. Lakra A and D K Yadav, "Multi-Objective Tasks Scheduling Algorithms for
Cloud Computing Throughput Optimization", Procedia, Computer Science,
48(2015), pp 107-113.
26. Hamad S and Omara F, "Genetic-Based Task Scheduling Algorithm in Cloud
Computing Environment", International Journal of Advanced Computer Science
and Applications, 7(4), 2016, pp 550-556.
27. Madni S, Latif A and Usman M J," Performance comparison of heuristic
algorithms for task scheduling in IaaS cloud computing environment", Plosone,
Vol 3(2), 2017, pp 36-47.
28. R. N. Calheiros, R. Ranjan, A. Beloglazov A, C. A. F. D. Rose, R. Buyya,
“CloudSim: A Toolkit for Modeling and Simulation of Cloud Computing
Environments and Evaluation of Resource Provisioning Algorithms”, Software:
Practice and Experience, vol. 41(1), pp. 23–50, 2017.
29. Weiwei Lin, Chen Liang, James Z. Wang, Rajkumar Buyya, “Bandwidth-aware
divisible task scheduling for cloud computing”, Software Practice and Experience,
vol. 44, pp. 163–174, 2014.
30. Hong Xu, Baochun Li, “Anchor: A versatile and Efficient Framework for
Resource Management in the Cloud”, IEEE Transactions on Parallel and
Distributed Systems, vol. 24, pp. 1066-1076, June. 2019.