An Enhanced Workflow Scheduling Algorithm in Cloud Computing

Nora Almezeini and Alaaeldin Hafez

College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia

Keywords: Cloud Computing, Scheduling Algorithm, Fault Tolerant, Pricing Models, Workflow.

Abstract: Cloud Computing has gained high attention by provisioning resources and software as a service.

Throughout the years, the number of users of clouds is increasing and thus will increase the number of tasks

and load in the cloud. Therefore, scheduling tasks efficiently and dynamically is a critical problem to be

solved. There

are many scheduling algorithms that are used in cloud computing but most of them are

concentrating on minimizing time and cost and some of them concentrate on increasing fault tolerance.

However, very few scheduling algorithms that considers time, cost, and fault tolerance at the same time.

Moreover, Considering pricing models in developing scheduling algorithms to provide cost-effective fault

tolerant techniques is still in its infancy. Therefore, analysing the impact of the different pricing models on

scheduling algorithm will lead to choosing the right pricing model that will not affect the cost. This paper

proposes developing a scheduling algorithm that combines these features to provide an efficient mapping of

tasks and improve Quality of Service (QoS).

1 INTRODUCTION

Cloud computing is considered as a distributed

system that offers services to the Internet users

through service providers such as Amazon, Google,

Apple, Microsoft, and others. Cloud computing uses

Internet technologies to offer elastic services that

support dynamic access to the computing resources

and support variable workloads.

However, cloud computing still requires more

scientific research across a variety of different topics

in order to gain its full benefits. One of the important

topics that need to be researched is the performance

efficiency of scheduling where workflow scheduling

focuses on efficiently mapping tasks to appropriate

resources. Finding optimal solution in cloud

computing is considered as NP-Complete Problem

(Kalra and Singh, 2015). Each scheduling algorithm

based on one or more strategy. The most important

strategies or objectives commonly used are time,

cost, energy, Quality of Service (QoS), and fault

tolerance (Chandrashekar, 2015). There are many

scheduling algorithms have been applied in cloud

systems such as Min-Min, Max-Min, Genetic

Algorithm, Particle Swarm Optimization (PSO), and

Heterogeneous Earliest Finish Time (HEFT)

(Rahman et al., 2013, Chaudhary and Kumar, 2014b,

Devipriya and Ramesh, 2013). Those algorithms and

others have been concentrating on minimizing the

overall completion time of the schedule (makespan)

and the cost. However, very few algorithms have

taken fault tolerance in consideration at the same

time. (Chandrashekar, 2015)

Developing a fault tolerant system is important

to provide proper and continuous work even if

failures are detected. There are two types of fault

tolerant techniques that can be implemented in the

cloud computing services: Proactive techniques that

predict failures and prevent them by replacing

suspected components with working components,

and Reactive techniques try to reduce the impact of

failure after it occurs and recover it. (Nazari

Cheraghlou et al., 2015, Sarmila et al., 2015)

Applying fault tolerant techniques in the system

in order to detect and recover any failure attack the

system is very important. However, it costs a lot. For

example, replication technique needs more resources

that will increase the cost of scheduling. Also,

considering pricing models in developing scheduling

algorithms to provide cost-effective fault tolerant

techniques is still in its infancy. Therefore, it is

important to analyse the impact of the fault tolerant

technique on cost by considering the pricing models

that will lead to a cost-effective fault tolerance.

(Chandrashekar, 2015)

Pricing in cloud environment is a very important

concept since cloud providers are mostly concerns

on maximizing revenue and profitability while the

Almezeini, N. and Hafez, A.

An Enhanced Workﬂow Scheduling Algorithm in Cloud Computing.

In Proceedings of the 6th International Conference on Cloud Computing and Services Science (CLOSER 2016) - Volume 2, pages 67-73

ISBN: 978-989-758-182-3

end users are more focused on cost-effectiveness in

addition to QoS. It is challenging to choose the right

pricing model that satisfies all parties. There are two

general pricing models in cloud computing: Fixed

Pricing Model (Pay-per-Use) where the consumers

are charged once for a computing capacity, and

Dynamic Pricing Model to provision fair resource

allocation with service differentiation.(Arshad et al.,

2015, AlRoomi et al., 2013)

The aim of this research is to show that it is

necessary to produce an enhanced workflow

scheduling system that combines the previous main

goals: reducing makespan and cost, increasing the

fault tolerance and robustness, and applying the

proper pricing model. Thus it will satisfy the

provider and consumer in the same time.

The rest of this short paper will talk about the

workflow scheduling and the most popular

scheduling algorithms used in clouds. Then, a brief

description of the important techniques of fault

tolerant that can be used in cloud computing. Also,

the pricing models of cloud computing will be

described in section 4. Then, research objectives and

methodology of the proposed research are presented

in section 5 and 6.

2 WORKFLOW SCHEDULING

Workflow scheduling assigns tasks based on their

dependencies to shared resources that are controlled

by a workflow scheduler (Kalra and Singh, 2015).

The aim of workflow scheduling is to obtain the

desired QoS. Workflow scheduling has a lot of

advantages. It increases computing performance and

throughput which increase the user satisfaction and

reduces the execution cost and time. However, it is

important to map the tasks efficiently to given

resources in order to provide high QoS to users.

Therefore, workflow scheduling is considered a

Non-deterministic polynomial (NP)-Complete

problem (Kalra and Singh, 2015, Chandrashekar,

2015).

There are many workflow scheduling

methodologies and techniques are used to map tasks

to resources which can be classified into two types:

Heuristics and Metaheuristics. Heuristic techniques

are based on exhaustive search to provide an

approximate optimal solution although the operating

cost and complexity of generating schedules is very

high. Examples of heuristic techniques are Min-Min

and Max-Min algorithms. On the other hand,

metaheuristic techniques are able to solve large and

complex problems effectively and efficiently in

reasonable time. Some of the well known

metaheuristic techniques are Genetic Algorithms

(GA), and Ant Colony Algorithms (ACO).(Kalra

and Singh, 2015, Chandrashekar, 2015)

Each scheduling algorithm should be based on

one or more strategies. The most important

strategies or objectives commonly used are time,

cost, energy, QoS, and fault tolerance

(Chandrashekar, 2015). Moreover, workflow

scheduling has two types of planning schemes: static

and dynamic (Chandrashekar, 2015, Devipriya and

Ramesh, 2013). If the number of tasks is known

beforehand, static scheme is applied so the tasks are

mapped at the compile time. On the other hand, the

dynamic scheme is applied when the tasks are

arrived in a dynamic manner so some assumptions

are provided before execution and scheduling

decisions are made just in time.

Famous cloud providers are heavily reliant on

big data analytics such as Google that developed

Map-Reduce framework for processing big data

workflows. Also, Yahoo and Facebook use Hadoop

which is the open source implementation of Map-

Reduce framework. Hadoop has four choices of

scheduling algorithms: FIFO, Fair scheduler,

Capacity scheduler and Dynamic scheduler.

Moreover, IBM blue cloud is based on Xen and

Hadoop clusters. Amazon EC2 uses FIFO, default

algorithm in Hadoop. (Wang et al., 2014,

Chandrashekar, 2015)

2.1 Workflow Scheduling Algorithms

2.1.1 Min-Min Scheduling Algorithm

Min-Min scheduling algorithm was proposed (

Rahman et al., 2013) for scheduling tasks in grid

projects. It is simple and fast and works as the base

for most of cloud scheduling algorithms. It provides

better performance by scheduling the task with

minimum size to the resource that has the minimum

completion time (MCT) for all unmapped tasks

(Chaudhary and Kumar, 2014b). Finally, this task is

removed from the set of unmapped tasks and the

process is repeated again by Min-Min algorithm till

every task is assigned.

However, the performance of this algorithm is

not perfect since it schedules the small tasks at first

(Devipriya and Ramesh, 2013). Also, it works well

if the number of smaller tasks is greater than the

number of larger tasks. Min-Min uses a single

resource; therefore, it is unable to execute tasks

concurrently.

CLOSER 2016 - 6th International Conference on Cloud Computing and Services Science

QoS Guided Min-Min Algorithm (Chaudhary and

Kumar, 2014a) is a Min-Min algorithm with Quality

of Service (QoS) constraints added to it such as

bandwidth, time, and memory. Tasks with high QoS

request parameters are assigned to resources first. In

this, resources are utilized in higher rate and they are

scaled elastically in execution.

Segmented Min-Min Scheduling Algorithm

(Chaudhary and Kumar, 2014a) where the tasks are

sorted based on the Estimated Time to Complete

(ETC). Then it creates segments with the equal size

using task list partitioning scheme. Segments with

larger tasks are scheduled and executed before the

smaller tasks. In each segment, Min-Min algorithm

is applied to assign tasks to resources.

Double Min-Min Algorithm (Chaudhary and

Kumar, 2014a, Kong et al., 2011) reselects the task

that is greater than mean Completion Time and then

schedules the reselected task again using Min-Min

algorithm. This will provide better load balancing

and resource utilization.

2.1.2 Max-Min Scheduling Algorithm

Max-Min scheduling algorithm is very similar to

Min-Min except that Max-Min selects the task that

has the maximum MCT and assign it to the

resources having minimum execution time (Rahman

et al., 2013, Chaudhary and Kumar, 2014b,

Devipriya and Ramesh, 2013) . This algorithm gives

the priority to large tasks rather than small ones and

executes many small tasks while executing the large

task concurrently. Therefore, Max-Min performs

better than Min-Min algorithm if the number of

short tasks is more than long tasks (Devipriya and

Ramesh, 2013).

Many improvements have been applied to these

algorithms in order to minimize their drawbacks

especially that they are not useful for large scale

distributed systems. (Devipriya and Ramesh, 2013)

has improved the Max-Min algorithm by assigning

the task with maximum execution time to the

resource that has minimum completion time instead

of assigning task with maximum completion time to

the resource which produce minimum execution

time in the original Max-Min algorithm. The goal is

to assign the largest task to the slowest resource in

order to give the opportunity to small tasks to be

finished by high speed resources concurrently. This

improved algorithm has reduced the makespan.

However, it concerned only with the number of

resources and tasks.

In (Kaur, 2014), the authors have improved the

Max-Min algorithm as in (Devipriya and Ramesh,

2013) and combine it dynamically with Ant Colony

algorithm as hybrid approach. The improved Max-

Min algorithm produces an optimal solution in the

initial stage, but it reduces after some time.

However, the searching speed during the early stage

in the Ant Colony algorithm is very slow due to the

lacking of pheromones, and then the speed of

optimal solution increases quickly after pheromones

reach a certain degree. Therefore, the aim of the

integration in (Kaur, 2014) is to benefit from Max-

Min algorithm in the initial stage and then get the

optimal solution by Ant Colony algorithm in last

stage.

Resource Aware Scheduling Algorithm (RASA)

(Devipriya and Ramesh, 2013, Chaudhary and

Kumar, 2014a) is a hybrid algorithm based on Min-

Min and Max-Min algorithms. They are applied

alternatively in order to avoid their main drawbacks.

This algorithm calculates the completion time of

each task on the available resources and then applies

Max-Min and Min-Min algorithms alternatively to

take advantage of both of them and avoid their

drawbacks.

2.1.3 Genetic Algorithm

Genetic Algorithm is a metaheuristic technique that

provides useful solution to optimization problems by

applying the principles of evolution. A Genetic

Algorithm depends on two techniques to be effective

(Rahman et al., 2013):

 Exploitation: exploit the best solution from past

results.

 Exploration: explore the new areas of solution

space.

Genetic Algorithm begins by initializing a

population with random candidate solutions called

individuals. Each individual is evaluated by a fitness

function which can be different according to the

given optimization objective. Then a proportion of

population is selected to reproduce a new

generation. After that, two main genetic operators

are used to generate new generation population.

These two operators are: crossover and mutation.

(Rahman et al., 2013, Kumar and Verma, 2012)

Using genetic algorithms in scheduling is a

powerful approach since they provide better

solutions with the increase of population size and

number of generations. However, the random

generation of initial population leads to schedules

that are not so much fit, so when these schedules are

mutated with each other, there are a very low

probability to produce better child than themselves.

Therefore, many researches have worked on

An Enhanced Workﬂow Scheduling Algorithm in Cloud Computing

improving genetic algorithms especially in the early

steps in order to increase the performance.

In (Kumar and Verma, 2012), the authors have

improved genetic algorithm by using the Min-Min

and Max-Min algorithms in generating initial

population. This will provide better initial

population and better solutions than initializing

population randomly in standard genetic algorithms.

(Liu et al., 2014) proposed an algorithm that

combines between genetic algorithm and Ant

Colony algorithm. They benefit from the strong

global search capability of genetic algorithms in the

early stages of Ant Colony algorithm to generate the

initial population then convert it to initial pheromone

of ACO to provide the optimal solution. This

integration of global search capability and high

accuracy shows the good performance of task

scheduling and load balancing.

(Wang et al., 2014) have proposed a scheduling

algorithm through improved genetic algorithm in

order to minimize the makespan and load balancing.

They chose greedy algorithm to initialize the

population. This proposed algorithm showed better

performance in load balancing than genetic

algorithm.

2.1.4 Other Scheduling Algorithms

Particle Swarm Optimization (PSO) based Heuristic

(Pandey et al., 2010) is a scheduling algorithm that

considers computation cost and data transmission

cost to schedule applications to cloud resources. The

evaluation results show that PSO can obtain greater

cost savings and good workload distribution to

resources in the cloud.

Heterogeneous Earliest Finish Time Algorithm

(HEFT) (Chaudhary and Kumar, 2014a, Wieczorek

et al., 2005) determines the average execution time

for each task and also the average communication

time between the resource of two tasks. After that, it

orders tasks by rank function so the task with a

higher rank value is given a higher priority. So the

tasks are scheduled based on the priority order and

every task is assigned to resources that complete it at

earliest time.

Scalable Heterogeneous Earliest Finish Time

Algorithm (SHEFT)(Chaudhary and Kumar, 2014a,

Hirales-Carbajal et al., 2012) works similar to HEFT

in addition to scaling resources elastically at

runtime. Therefore, it obtains an optimized

execution time.

Round Robin Algorithm (Chaudhary and Kumar,

2014a, Chaudhary and Singh Chhillar, 2013) is a

static algorithm that determines a time slot for each

task, and suspends the task if it is not completed

during this time slot. Once all the tasks have finished

their time slots the first uncompleted task will get

again another allocation and the cycle will repeat.

However, this will overload the nodes in some times

and under load at other times. Therefore, the

Weighted Round Robin Algorithm (Chaudhary and

Kumar, 2014a, Chaudhary and Singh Chhillar, 2013)

assigns weights in order to each task so the tasks are

allocated to resources based on their weights and

time slots for optimal utilization of resources.

3 FAULT TOLERANT

MECHANISMS IN CLOUDS

Fault is a defect that affects the system and changes

its status to be unable to continue to work. In

general, there are three types of faults (Sarmila et al.,

2015, Nazari Cheraghlou et al., 2015):

• Transient: a fault that vanished when it fixed.

• Intermittent: a fault that could appears again

and again.

• Permanent: a fault that cannot be fixed.

Faults in cloud computing are different based on

computing resources. (Kumar et al., 2015) have

indicated that fault in cloud computing could be

Network Fault, Physical Fault, Processor Fault, and

Service Expiry Fault. Moreover, (Chandrashekar,

2015) mentioned that faults could be variations in

performance of resources, or unavailable files.

It is obvious that developing a fault tolerant

system is important to provide proper and

continuous work even if failures are detected. There

are different fault tolerant techniques that can be

implemented in the cloud computing services. These

techniques can be classified into two categories:

Proactive and Reactive techniques. Proactive

techniques avoid recovery after the occurrence of

fault. It predicts failures and prevents them by

replacing suspected components with working

components. This technique is more efficient than

reactive but not always the predictions are accurate.

Reactive techniques try to reduce the impact of

failure after it occurs and recover it. It is more

reliable. Therefore, it is mostly used. (Sarmila et al.,

2015, Nazari Cheraghlou et al., 2015)

3.1 Proactive Fault Tolerant

Techniques

Self-heading: this technique handles the failed

instances of an application on multiple virtual

CLOSER 2016 - 6th International Conference on Cloud Computing and Services Science

machines, since multiple instances of the application

are running on multiple VMs.

Preemptive Migration: this technique is based on

monitoring and analysing the application

continuously according to the feedback-loop

mechanism.

3.2 Reactive Fault Tolerant Techniques

Replication: tasks are replicated on different

resources in order to be able to recover quickly from

failures.

Resubmission: this technique resubmits the failed

task to alleviate failures.

Check Point/Restart: while the application is

running, it makes checkpoints in different parts. This

technique restarts the application at the last

checkpoint achieved when a failure occurs.

Rescue Workflow: this technique ignores the

failed tasks and allows the workflow to continue

until no more forward steps can be made and

becomes impossible to continue without handling

the failures.

User Defined Exception Handling: this technique

allows the user to specify a predefined action for

particular failures in a workflow.

Job Migration: when a failure occurs, this

technique will send the task to other similar virtual

machine.

4 PRICING MODELS IN CLOUDS

Cloud providers are mostly concerns on maximizing

revenue and profitability which can be employed by

several pricing models. On the other hand, end users

are more focused on Quality of Service (QoS),

availability and usability of resources, and cost-

effectiveness. It is seriously challenging cloud

providers to keep balance between these two

important purposes and choose the right pricing

model that satisfies all parties.(Arshad et al., 2015,

AlRoomi et al., 2013)

Pricing models in cloud computing can be

classified into two general models:

4.1 Fixed Pricing (Pay-Per-Use)

Cloud providers mainly used this approach in order

to provision their services to consumers. Consumers

are charged once for a computing capacity. This

strict model has some drawbacks since the user may

not use all the resources he charged for and also the

provider will not consider the QoS and other

satisfaction measures after charging the user.

Examples of cloud providers that employ this model

is Amazon on-demand instances and Google App

Engine.(AlRoomi et al., 2013)

Pay per use fixed pricing model is suitable for

IaaS and PaaS. It has two sub categories: Pay for

resources and Subscription. In Pay for resources

technique, users will be charged for using storage or

bandwidth size of resource. While in Subscription

technique, the user will subscribe to a service

provider with a fixed price per unit for a long

time.(Arshad et al., 2015)

4.2 Dynamic Pricing

Whereas the nature of cloud environment is dynamic

and the number of cloud users increasing, vendors

have adopted the modification from fixed pricing to

dynamic pricing in order to provision fair resource

allocation with service differentiation. Amazon’s

spot instances pricing model was the first concept of

dynamic pricing model which profits from the

available unused resources in the data centres

executing the requests of on-demand

instances.(Arshad et al., 2015)

5 RESEARCH OBJECTIVES

Based on the brief literature review, it is obvious

that scheduling tasks efficiently and dynamically is a

critical problem to be solved. Throughout the years,

the number of tasks in the cloud system and

dependencies between tasks are increasing. This will

affect the makespan and the cost. Therefore, it is

necessary to develop a scheduling algorithm that

prevents these challenges and provide better

performance in time and cost.

Moreover, the system will be susceptible to

failures and performance variations. Thus, making

the workflow scheduling algorithm efficient and

fault tolerant is very important in order to overcome

the drawbacks of previous algorithms.

Pricing in cloud environment is a very important

concept. Considering pricing models in developing

scheduling algorithms to provide cost-effective fault

tolerant techniques is still in its infancy. Therefore,

analysing the impact of the different pricing models

on scheduling algorithm will lead to choosing the

right pricing model that will not affect the cost.

The goal of this research is to propose an

enhanced workflow scheduling system that

combines the main goals: reducing makespan and

cost, increasing the fault tolerance and robustness,

An Enhanced Workﬂow Scheduling Algorithm in Cloud Computing

and applying the proper pricing model. Thus it will

satisfy the provider and consumer in the same time.

6 METHODOLOGY

As a first step, the research methodology starts with

studying the domain of cloud computing and

techniques and mechanisms of cloud computing.

Following this step, we will investigate previous and

current studies and works related to our research. As

a major step towards achieving our aim, we will

develop a scheduling algorithm that improves the

performance of the current scheduling algorithms in

terms of reducing time and cost. Secondly, a proper

fault tolerant mechanism will be integrated to the

developed scheduling algorithm so that does not

affect the time and cost and increases the reliability

and robustness of the system. Finally, analysing the

impact of the different pricing models on the

proposed scheduling algorithm will lead to choosing

the right pricing model for the proposed algorithm in

which it will not affect the previous steps.

A simulator such as CloudSim will be used to

evaluate the performance of the proposed algorithm

since it is very difficult to conduct large scale

experiments on real cloud infrastructures as well as

it is time consuming and costly. The proposed

algorithm will be evaluated first in terms of time and

cost. The results will be compared with previous

algorithms that consider the same goal. Then we will

evaluate the algorithm in terms of fault tolerant and

compare the results with the algorithms that apply

the same techniques. After that, we will analyse the

contribution of the selected pricing model in

reducing the cost of applying the fault tolerant

mechanism in the proposed algorithm.

Finally, the proposed algorithm will be evaluated

as whole and analysed in terms if it achieved the

desired objective and if it increased the efficiency of

workflow scheduling.

7 CONCLUSIONS

It is explicit that mapping tasks efficiently to given

resources in order to ensure Quality of Service

(QoS) is a major challenge. If this is not achieved,

the user will hesitate to join the cloud and pay.

Minimizing makespan and cost, increasing fault

tolerance, and choosing the proper pricing model are

very important objectives that will improve the QoS.

Therefore, combining the features in a workflow

scheduling is necessary to provide efficiency and

gain satisfaction from providers and consumers.

ACKNOWLEDGEMENTS

This research project is supported by a grant from

the Deanship of Graduate Studies, King Saud

University, Saudi Arabia.

REFERENCES

Alroomi, M., Alebrahim, S., Buqrais, S. & Ahmad, I.

2013. Cloud Computing Pricing Models: A Survey.

International Journal Of Grid And Distributed

Computing, 6, 93-106.

Arshad, S., Ullah, S., Khan, S. A., Awan, M. D. & Khayal,

M. S. H. A Survey Of Cloud Computing Variable

Pricing Models. Evaluation Of Novel Approaches To

Software Engineering (Enase), 2015 International

Conference On, 29-30 April 2015 2015. 27-32.

Chandrashekar, D. P. 2015. Robust And Fault-Tolerant

Scheduling For Scientific Workflows In Cloud

Computing Environments.

Chaudhary, D. & Kumar, B. 2014a. An Analysis Of The

Load Scheduling Algorithms In The Cloud Computing

Environment: A Survey. 2014 9th International

Conference On Industrial And Information Systems

(Iciis).

Chaudhary, D. & Kumar, B. 2014b. Analytical Study Of

Load Scheduling Algorithms In Cloud Computing.

2014 International Conference On Parallel,

Distributed And Grid Computing.

Chaudhary, D. & Singh Chhillar, R. 2013. A New Load

Balancing Technique For Virtual Machine Cloud

Computing Environment. International Journal Of

Computer Applications, 69, 37-40.

Devipriya, S. & Ramesh, C. 2013. Improved Max-Min

Heuristic Model For Task Scheduling In Cloud. 2013

International Conference On Green Computing,

Communication And Conservation Of Energy (Icgce).

Hirales-Carbajal, A., Tchernykh, A., Yahyapour, R.,

González-García, J. L., Röblitz, T. & Ramírez-

Alcaraz, J. M. 2012. Multiple Workflow Scheduling

Strategies With User Run Time Estimates On A Grid.

Journal Of Grid Computing, 10, 325-346.

Kalra, M. & Singh, S. 2015. A Review Of Metaheuristic

Scheduling Techniques In Cloud Computing. Egyptian

Informatics Journal.

Kaur, R. 2014. Hybrid Improved Max Min Ant Algorithm

For Load Balancing In Cloud. In: Ghumman, N. (Ed.)

International Conference On Communication,

Computin G & Systems.

Kong, X., Lin, C., Jiang, Y., Yan, W. & Chu, X. 2011.

Efficient Dynamic Task Scheduling In Virtualized

Data Centers With Fuzzy Prediction. Journal Of

Network And Computer Applications, 34, 1068-1077.

CLOSER 2016 - 6th International Conference on Cloud Computing and Services Science

Kumar, P. & Verma, A. 2012. Independent Task

Scheduling In Cloud Computing By Improved Genetic

Algorithm. International Journal Of Advanced

Research In Computer Science And Software

Engineering (Ijarcsse), 2, 111-114.

Kumar, S., Singh Rana, D. & Chandra Dimri, S. 2015.

Fault Tolerance And Load Balancing Algorithm In

Cloud Computing: A Survey. International Journal Of

Advanced Research In Computer And Communication

Engineering, 4, 92-96.

Liu, C.-Y., Zou, C.-M. & Wu, P. 2014. A Task Scheduling

Algorithm Based On Genetic Algorithm And Ant

Colony Optimization In Cloud Computing. 2014 13th

International Symposium On Distributed Computing

And Applications To Business, Engineering And

Science.

Nazari Cheraghlou, M., Khadem-Zadeh, A. & Haghparast,

M. 2015. A Survey Of Fault Tolerance Architecture In

Cloud Computing. Journal Of Network And Computer

Applications.

Pandey, S., Wu, L., Guru, S. M. & Buyya, R. 2010. A

Particle Swarm Optimization-Based Heuristic For

Scheduling Workflow Applications In Cloud

Computing Environments. 2010 24th Ieee

International Conference On Advanced Information

Networking And Applications.

Rahman, M., Hassan, R., Ranjan, R. & Buyya, R. 2013.

Adaptive Workflow Scheduling For Dynamic Grid

And Cloud Computing Environment. Concurrency

And Computation: Practice And Experience, 25, 1816-

1842.

Sarmila, G. P., Gnanambigai, N. & Dinadayalan, P. 2015.

Survey On Fault Tolerant — Load Balancing

Algorithmsin Cloud Computing. 2015 2nd

International Conference On Electronics And

Communication Systems (Icecs).

Wang, T., Liu, Z., Chen, Y., Xu, Y. & Dai, X. 2014. Load

Balancing Task Scheduling Based On Genetic

Algorithm In Cloud Computing. 2014 Ieee 12th

International Conference On Dependable, Autonomic

And Secure Computing.

Wieczorek, M., Prodan, R. & Fahringer, T. 2005.

Scheduling Of Scientific Workflows In The Askalon

Grid Environment. Acm Sigmod Record, 34.

An Enhanced Workﬂow Scheduling Algorithm in Cloud Computing