8 V May 2020

http://doi.org/10.22214/ijraset.2020.5369
 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue V May 2020- Available at www.ijraset.com

Task Scheduling in Cloud Computing: A Survey

Gopika Venu1, Vijayanand K S2
1
M. Tech. Student, Dept. of Information Technology, Government Engineering College Bartonhill, India
2
Associate Professor, Dept. of Information Technology, Government Engineering College Bartonhill, Kerala, India

Abstract: The on-demand availability of computer system resources such as data storage and computing power is cloud
computing. Scheduling is the method of allocating jobs onto resources in time. Scheduling increases the efficiency and
performance of cloud environment by maximizing the resource utilization. This scheduling process has to respect constraints
given by the jobs and the cloud providers. Ordering the tasks by scheduler along with maintaining the balance between Quality
of Service (QoS), fairness and efficiency of jobs is difficult. Scheduling algorithms are designed and implemented considering
some parameters like latency, cost, priority, etc. The aim of this paper is a study of various types of job scheduling algorithms
that provide efficient cloud services.
Keywords: Task scheduling, Cloud computing, on-Demand computing.

I. INTRODUCTION
Cloud is a trendy term for a network or remote servers that can be accessed via an Internet connection store and manage information.
Two types of models present such as deployment models and service models. The deployment model includes private, public and
hybrid cloud. Private Clouds are data centers architecture owned by a single company. E.g. IBM’s Blue Cloud, Sun Cloud, Window
Azure, etc. Public Clouds are basically the Internet. To make resource available to general people, service provider use internet. E.g.
Gmail, Office 365, Drop box, etc. In case of hybrid clouds, for instance during peak periods individual applications or portion of
applications can be migrate to the public cloud. The service model includes Software as a Service (SaaS) such as Drop box, Google
Apps, Platform as a Service (PaaS) such as Google Compute Engine, Apache Stratos and Infrastructure as a Service (IaaS) such as
Virtual Machine, Storage, Servers.
Cloud provides on demand computational resources in the form of virtual machines .These Virtual Machines are deployed in a
cloud provider’s data centers. The computational resources are shared among different cloud consumers who pay for the service
accessed as per the usage. Allocation of resources and proper scheduling has a considerable impact on the performance and
efficiency of the system. The main goal of cloud computing is to provide efficient access to remote and geographically distributed
resources. An efficient scheduling is a key to manage the access to different resources, load balancing as well as resource allocation.
Different types of resource scheduling algorithms are available in cloud computing based on certain parameters like time, cost, and
performance, utilization of resources throughput, bandwidth, resource availability, physical distances and priority.
The paper is organized as follows. Section II includes Scheduling in Cloud Computing. Section III describes literature survey on
scheduling algorithms used in cloud computing. And section IV includes conclusion.

II. SCHEDULING IN CLOUD COMPUTING

Scheduling is the allocation of various jobs to available resources. The task is a minimum computational unit to run on a node or
resource. Job is a computational activity made up of several tasks that could require different processing capabilities and could have
different resource requirements such as CPU, number of nodes, memory, etc. Each job may have various parameters such as desired
completion time often called the deadline, required data, expected execution time, job priority, etc. A resource is something that is
required to carry out an operation. For e.g., a processor for data processing, a data storage device or a network link for data
transporting, etc. Scheduling problem is related to two types of users cloud consumers as well as cloud providers. Cloud consumer
wants to execute their jobs for solving problem of varying size and complexity. Cloud providers will contribute resources for
executing consumer jobs. Cloud consumer will benefit by selecting and aggregating resources wisely while cloud provider benefit
by maximizing resource utilization. The problem of mapping tasks on resources belongs to a class of problems known as NP hard
problems. These are problems for which no known algorithms are able to generate the optimal solution within polynomial time. So
solutions based on exhaustive search are impractical. Overhead of generating schedules is very high. Scheduling decisions must be
made in the shortest time possible in clouds because multiple users are competing for the time and resources, so metaheuristic
algorithms such as genetic algorithm, PCO etc are used.

©IJRASET: All Rights are Reserved 2258

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue V May 2020- Available at www.ijraset.com

A. Framework Of Task Scheduling In Cloud Computing Environment

There is a framework for task scheduling system in cloud computing. In Fig 1, the users are accessing the cloud environment
through internet. The cloud part shows how the cloud is managing to serve various requests given by the consumers.

Fig1: Framework of task scheduling in cloud computing environment

1) SLA Monitor: The customer first submits the service request. Then the SLA monitor checks the submitted request for QoS
requirements before determining whether to accept or reject the request. SLA Monitor is responsible to monitor the progress of
the submitted job and if there are any violations, corrective actions has to be taken immediately.
2) Resource Discovery And Monitoring: Resource discovery may be described basically as the task in which the provider should
find appropriate resources in order to comply with incoming consumer requests. The key features of cloud computing is
capability of receiving and releasing resources on demand, therefore monitoring of resources has to be continuous.
3) Task Scheduling: The input of task scheduling algorithm is normally an abstract model .This abstract model defines tasks
without specifying the physical location of resources on which the tasks are executed.
4) Reschedule: When a task can’t be completed due to processor failure or other problems then uncompleted task would be
rescheduled in the next computation.
5) Scheduling Optimizer: After acquiring information about available resources in the cloud a set of appropriate candidate is
highlighted .The resource selection mechanism selects the candidate solution that fulfills all requirements and optimizes the
usage of the infrastructure .The Resource selection may be done using an Optimization Algorithm. Many Optimization
strategies may be used from simple and well known techniques such as simple metaheuristic algorithms. Genetic Algorithm,
Ant Colony and Particle Swarm Optimization for clouds are examples.
6) Advanced Resource Reservation Monitor: Advanced resource reservation monitor provides QoS guarantees for accessing
various resources across data centre. Users are able to secure resources required in the future. This is important to ensure
successful completion of time critical applications such as real time and workflow applications for parallel applications
requiring a number of processors to run. Provider can predict future demand and usage more accurately.
7) Data Centers and VMs: Data centers are the infrastructure or hardware part where all the physical servers present. Physical
servers are not used as viable users. Through virtualization they are converted into different VMs and users jobs are going to
run on these VMs in scheduling tasks.VM manager manages the VMs.

©IJRASET: All Rights are Reserved 2259

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue V May 2020- Available at www.ijraset.com

B. Different Types Of Scheduling

1) Static vs. Dynamic Scheduling: Jobs are pre scheduled in Static scheduling. Information regarding all resources as well as the
entire task in the application is assumed to be available by the time the application is scheduled. The main drawback is no job
failure and resources failure is assumed. In dynamic scheduling, jobs are dynamically available for scheduling overtime by the
scheduler with no issues to be able of determining runtime in advance. The Dynamics of job execution which refers to the
situation when job execution could fail due to some resource failure or job execution could be stopped due to the arrival in the
system of high priority jobs in the case of preemptive mode is considered. The Dynamics of resources in which workload on
resource can change continuously overtime
2) Centralized, Hierarchical vs. Distributed Scheduling: Centralized and decentralized scheduling differ in the control of the
resources and knowledge of the overall system. In the case of centralized scheduling there is more control on resources. A
scheduler has knowledge of the system by monitoring of the resource state. The main advantages are ease of implementation,
efficiency and more control and monitoring of resources. The main disadvantage is lacks scalability. Hierarchical scheduling
allows one two co-ordinate different schedules at certain level. Schedulers at the lowest level in the hierarchy have knowledge
of the resources. The main disadvantage is lack of scalability and fault tolerance yet it is better and is more fault-tolerant than
centralized schedulers. In Decentralized or distributed scheduling, no central entity controlling the resources .Scheduling
decisions are shared by multiple distributed schedulers. This is less efficient than centralized schedulers.
3) Preemptive vs. Non Preemptive Scheduling: Preemptive scheduling criterion allows job to be interrupted during execution and a
job can be migrated to another resource leaving its originally allocated resource and unused to be available for other jobs it is
more helpful if there are constraints as priority to be considered. In case of Non preemptive, resources aren’t being allowed to
be reallocated until the running and scheduled job finish the execution.
4) Immediate vs. Batch Mode Scheduling: Immediate or online mode scheduler schedules recently arriving job as soon as it arrives
with no waiting for next time interval on available resources at the moment. Batch or offline mode scheduler holds arriving jobs
as group of problems to be solved over successive time intervals. This will helps in better mapping in job for suitable resource
depending on its characteristics.
5) Independent vs. Workflow Scheduling: In independent scheduling, tasks can be run independently. In the case of workflow
scheduling tasks are interdependent. Dependency means there are precedence orders existing in tasks .That is a task cannot start
until all its parent are done. Workflows are represented by directed acyclic graph notation. Each task can start its execution only
when all preceding tasks in DAG are already finished. In DAG nodes are tasks and connections or edges are dependencies
between them.

C. Scheduling Objectives
An objective function is designed for scheduling algorithm. The main aim is to minimize this objective function or maximize this
objective function according to the criterion specified by the user. In traditional approaches a single objective function is considered
but now multicriterion objective function is created. In this, simultaneously two or more parameters are considered. On the basis of
users it can be Application Centric. For example the parameters makespan and economic cost can be application centric. The
designed algorithm should minimize either makespan or economic cost or both. On the basis of providers it can be Resource Centric.
For example the parameters resource utilization and economic profit can be resource centric. The designed algorithm should
maximize either resource utilization or economic profit or both.

III. LITERATURE SURVEY

A. Various Existing Task Scheduling Algorithms
1) First Come First Serve (FCFS): The incoming task looks for the queue where the waiting time is the smallest. The queue is
managed by FIFO mechanism (first come first out). In FCFS, place each incoming Task at the end of the service queue. The
first task in the queue is assigned to VM when it is available. This algorithm is simple and easy to understand. The main
disadvantage is large waiting time. The short jobs at the back of a queue will wait until large task in the front of queue is
finished. It is based on single criterion for scheduling. This algorithm is non preemptive.
2) Round Robin Algorithm: Round Robin Algorithm is a pre-emptive algorithm .This algorithm distributes the jobs on the
available VMs in a round form or cyclic manner, where the jobs are stored in a ring queue. Each job is given a quantum of time.
If it is not completed within its turn, it would be interrupted. Then it is stored back in the tail of the queue, and wait for its next
turn. The algorithm is repeated until each task in the queue is assigned to minimum one virtual machine. The pros of this

©IJRASET: All Rights are Reserved 2260

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue V May 2020- Available at www.ijraset.com

algorithm are no need for a preprocessing step to fetch the nominated VM, Distribute the load equally among VMs, Focuses on
fairness among the scheduled tasks, Jobs are executed in turn and never waiting for previous job to finish execution (starvation
free), The scheduler will not wait until all processing power of a VM is exhausted before it moves to next VM and at last, it is
based on a simple rule. The major cons of this algorithm are long jobs take longer time to complete execution. Servers may be
overloaded. Preemptive policies depend on the length of time slice and during short time slice this would cause many switching.
3) Modified round robin [1]: By dynamically taking into consideration the burst time for each incoming task entering the ready
queue, modified round robin does a simple improvement on the standard round robin algorithm. The time slicing process is
done on the basis of computing the average burst time of all the remaining waiting requests in the ready queue. For this purpose,
two registers are there, Sreg and Areg. The total burst time of all request in the ready queue are stored in Sreg. The Average
burst time by dividing the value of Sreg by the number of tasks residing in the ready queue is stored in the Areg. In the
beginning the first job is allocated to virtual machine and all its burst time is taken, then the time slice of each incoming request
is computed by the scheduler. When it is allocated to run on a virtual machine, each task would run for a time period that is
same as the time slice granted to it by modified Round Robin, when it entered the queue. When the time slice is over the task is
either removed from the ready queue or joins the ready queue again standing at the back of the queue. The scheduler adjusts the
values of the registers by subtracting the burst time of the removed task from Sreg and adding the value of the new joined tasks
to Areg.
4) Shortest Job First (SJF): Shortest job first scheduling (SJN) is used for ordering a set of tasks by putting the shorter task in the
front of the queue and longer tasks at the end of the queue. This algorithm reduces the waiting time of the short jobs and
increases the waiting time of the long jobs thereby reducing average waiting time .This can result in starvation for longer jobs
when there are a large number of small jobs.
5) Traditional Min-Min Heuristic Algorithm [2]: This algorithm is based on the concept of Minimum Completion Time. In Min-
Min algorithm, for each task determine its minimum completion time over all machines. For all tasks find the minimum
completion time. Assign the task to the machine that gives this completion time. Then iterate till all the tasks are scheduled. The
main advantage of this algorithm is smaller makespan, since tasks are scheduled on the fastest machines where they are
completed earlier. The algorithm is operative for the task scheduling in cloud computing. This algorithm will increase the
throughput. The disadvantages of this algorithm is, it increases the total completion time of all the tasks and hence increases the
makespan. The long tasks have to wait for smaller tasks to end their execution and the load is unbalanced.
6) QoS Guided Min-Min Heuristic Algorithm [3]: QoS guided Min-min takes bandwidth requirement of tasks into account. Jobs
requiring higher bandwidth are scheduled prior to others. Hence, if bandwidth requisite of all tasks vary tremendously, QoS
guided Min-min is better. In this algorithm, tasks are classified under high and low bandwidth. Task required high bandwidth
are scheduled first.
7) Traditional Max-Min Heuristic Algorithm [4]: This algorithm is based on the concept of Maximum Completion Time. In Max-
min algorithm, for each task determine its minimum completion time over all the machines. Then find the maximum
completion time for all. Assign the task to the machine that gives this completion time. Iterate till all the tasks are scheduled.
This can reduce the waiting time of long tasks so they never starved. This can increase the utilization and the response time is
minimized. The makespan is reduced since smaller jobs are executed concurrently while other longer jobs are executed. The
main disadvantage of this algorithm is as it first selects the large tasks for execution the smaller tasks are delayed. This
algorithm is not effective in load balancing.
8) Improved Max Min Algorithm [5]: Improved Max Min is based on the expected execution time and not on the completion time.
Max-Min could execute multiple short tasks simultaneously while executing larger ones. In this scenario the total makespan is
decided by the execution of long task. In cases where Meta tasks contain tasks having different completion time and execution
time, the makespan is not decided by one of submitted tasks. Improved max-min task scheduling algorithm tries to minimize
waiting time of short jobs by assigning largest tasks to be executed by slower resources, small tasks are executed concurrently
on fastest resources to complete large number of tasks during finalizing minimum one large task on slower resource. Based on
these cases they proposed a substantial improvement of Max min algorithm which leads to increase of max min efficiency. The
suggested improvement increases the opportunity of simultaneous execution of tasks on resources. It works in two phases: The
first phase is task with maximum execution time is selected (Largest Task).The second phase is selected task is scheduled over
resource with minimum completion time (Slowest Resource).
9) Enhanced Max-min Task Scheduling Algorithm [6]: Sometimes largest task is too large compared to other tasks in Meta-task.
In those cases, overall makespan is increased because too large task is executed by slowest resource first while other tasks are

©IJRASET: All Rights are Reserved 2261

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue V May 2020- Available at www.ijraset.com

executed by faster resource or when there is a huge difference among slowest and fastest sources in terms of processing speed
or band width. Then the largest task is executed by the slowest resource which increases the makespan and load imbalance
across resources. Therefore rather than selecting largest task, if we select average or closest greater than average task then
overall makespan is reduced and also balance load across resources.
10) Minimum Completion Time Algorithm: The algorithm scans the available VMs to find the most suitable machine to assign a job.
The VM is selection is based on the minimum completion time. This minimum completion time is calculated by taking into
consideration the processing speed and the current workload on a machine. Therefore it is considered as successful heuristic
that could be implemented in Cloud Computing. The main disadvantage is process of assigning a task to certain machine with
minimum completion time is done in arbitrary order so each time a task is assigned to the fastest machine in the remaining
resources pool.
11) Suffrage heuristic: Various steps to be performed are:
a) For each task determine the difference between its minimum and second minimum completion time over all the machines.
b) For each task determine the difference between its minimum and second minimum completion time over all the machines.
c) Over all the tasks find the maximum suffrage.
d) Assign the task to the machine that has resulted in obtaining minimum completion time.
e) Iterate till all the tasks are scheduled.
12) Resource Awareness Scheduling Algorithm [7]: Resource Awareness Scheduling Algorithm is a hybrid algorithm. It is a
combination of Max-Min and Min- Min Algorithms. This algorithm is also known as RASA. RASA is based on the concept of
Completion Time of each task. In Resource Awareness Scheduling Algorithm, for each task Expected Completion time is
calculated. To schedule task, apply Max-min and Min-Min alternatively. If number of resources is even, apply Max-Min
strategy first otherwise Min-Min strategy.
13) Reliable RASA Scheduling Algorithm [8]: To provide required QoS to user a resource reliability parameter is added to standard
algorithms. Reliability plays a significant role in performance of grid. Sometimes performance in view of completion time is
high but reliability of resource might be low. It is important to allocate tasks to highly reliable resources because low resource
reliability indicates frequent resource failures. In this algorithm, resource reliability is provided in percentage. Resource
manager sets reliability criteria. Appropriate resources which fulfill the criteria are selected. The result after applying this
algorithm is risk associated with job failure reduced.
It works in three phases:
a) Expected Completion time is calculated for each task.
b) Average or closest greater than average task is selected.
c) Selected task is scheduled over resource with minimum completion time i.e., with slowest resource.
d) For each task determine the difference between its minimum and second minimum completion time over all the machines.
e) For each task determine the difference between its minimum and second minimum completion time over all the machines

14) Genetic Algorithm [9]: The Genetic Algorithm mainly works in eight phases:
a) Genetic Encoding: Two dimension coding is the coding of the population individual.
b) Genetic Decoding: The decoding scheme of the encoded chromosomes is, the first char in genetic encode is decoded directed as
the host resource. Then, the following chars are decoded as the tasks order that are scheduled on this host resource.
c) Initial Population Generation: First genetic individual of initial population is generated.
d) Fitness Function: To measure the quality of the population individual, the fitness function of the population is used. Deadline,
budgets etc can be considered.
e) Genetic Crossover: Obtain new individuals in the current population by combining and rearranging parts of the existing
individuals. A crossover probability is selected to bring a better population individual by combining two fittest individuals.
f) Genetic Mutation: Two genetic mutation operations such as exchange mutation and replace mutation present. To select randomly
a host and select two tasks on this host to exchange, exchange mutation is used. To reallocate available hosts to tasks in the
population individual, replace mutation is used.
g) Genetic Selection: The Roulette wheel is used to implement the population individual.
h) Genetic Termination: The termination condition of the genetic operation is setting the maximal iteration number.

©IJRASET: All Rights are Reserved 2262

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue V May 2020- Available at www.ijraset.com

Using GA for tasks scheduling, each job vector is represented by chromosomes and the positions in this vector are tasks. The
population shows various mappings for tasks to machines and GA performs heuristic search to find the optimal solution. The fitness
function measures the quality of solution. Usually algorithm imitates the mechanism of natural selection strategy which consists of
four steps Selection crossover mutation and evaluation.
The main disadvantages are complexity in computations and long time requirement. The accuracy of algorithm can be decreased by
trial/error.
B. Enhanced Methods For Task Scheduling Procedures In Cloud Environment
For making the scheduling process more efficient, various hybrid algorithms and methods are proposed.
1) Particle Swarm Optimization Algorithm: PSO is a type of meta- heuristics algorithms. This algorithm applies self-adaptive
global search for optimization. It starts with random initialization for position and velocity for the practices population. Looking
at the problem of task scheduling, tasks are considered as the particles and number of tasks in the workflow is the dimension of
these particles. Each dimension has a value that indicates the resource where the tasks workflow is heading. So the mapping
between tasks and resources is shown by a particle in PSO. Like GA each particle is evaluated using fitness function. The
traffic workload using PSO is balanced. As it could be used with any number of tasks and resources, this algorithm is scalable.
It can find near optimal solutions for mapping all tasks in the workflow to the set of available resources. Less use of
mathematical operators compared with GA and consequently less need for parameters tuning. This algorithm is Simple and
effective. This can be used in wide applications with little computation overhead compared to GA. The major disadvantages are,
it is easy to fall into local optimum in large search space and slow convergence.
2) Cloud Task Scheduling Based on Ant Colony Optimization [10] [11]: ACO mainly simulate the food searching behaviour of ant
colonies. While searching for food ants use kind of chemical for communicating with each other. This Chemical is called
pheromone. In the beginning ants search for their foods randomly, once they find a food source, they leave pheromone on the
path. Any ant can reach the food source by following the trail of pheromone. As this process continues, ants try to find the
shortest path as there is large amount of pheromones accumulated on the way. Some of the advantages of this algorithm are
they use positive feedback mechanism, inner parallelism and this algorithm is extensible. The disadvantages are stagnation
phenomenon, all individuals find the same solution, searching to a certain extent, unable to search for a solution space, making
the algorithm reach a local optimal solution.
ACO can be applied to any problem if it defines:
a) The problem representation allows ants to incrementally build or modify solutions.
b) A constraint satisfaction method which forces the construction of possible solutions.
c) A pheromone updating rule has to be there which shows how to modify trails on the edges of the graph.
d) A transition rule of the heuristic desirability and of pheromone trail.
3) A New Flower Pollination based Task Scheduling Algorithm in Cloud Environment [12]: For scheduling tasks in cloud
environment here uses nature-inspired algorithm called Flower pollination Algorithm. This task scheduling approach is
proposed in order to map tasks and resources in best optimized way, thus minimizing make-span as a consequence. Set of tasks
of different completion times and set of resources of different processing powers are considered. In this algorithm, flowers are
single solutions. An initial population is randomly generated. To decide between local pollination and global pollination a
Switch probability is considered. For local pollination, the probability value is more. Global pollination and local pollination
can be calculated using the single solutions. If switching probability is less than a random number generated then global
pollination is applied, otherwise local pollination is applied. Then evaluate new solutions as per fitness Function. If new
solutions are better than existing ones then update them in population, otherwise ignore.
4) A Hybrid Bio-Inspired Algorithm for Scheduling and Resource Management in Cloud Environment [13]: Cloud receives
client’s tasks in a random manner. Allocation of resources to these tasks must be handled in an intelligent way. In this method,
first allocate the tasks to the virtual machines in an efficient manner. This is done using Modified Particle Swarm Optimization
algorithm. Then allocation or management of resources, as demanded by the tasks, is handled by the Hybrid Bio-Inspired
algorithm. Hybrid Bio-Inspired algorithm is a new hybrid approach. This incorporate two existing approaches such as Modified
Particle Swarm Optimization (MPSO) and Modified Cat Swarm Optimization (MCSO).This approach focuses on two goals.
First one is to provide efficient load balancing in task scheduling by building enhanced PSO algorithm. Second one is to
provide dynamic resource allocation and resource management by building hybrid approach using MPSO and MCSO
algorithms.

©IJRASET: All Rights are Reserved 2263

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue V May 2020- Available at www.ijraset.com

5) Monkey Search Algorithm for Task Scheduling in Cloud IaaS [14]: This Monkey Search Algorithm is a dynamic learning
inspired task allocation algorithm. This algorithm is used to minimize the overall completion time and to improve resource
utilization. Existing algorithms are mainly static or dynamic in nature. They may suffer from network traffic delay which leads
to request failure. Network aware monkey search algorithm with fine execution time, with lease execution cost and network
delay will provided better QoS than existing static, dynamic and learning based algorithms. This monkey search algorithm for
task allocation is divided into 4 phases such as Initialization, Climb, Watch and Jump and Somersault. In initialization, initial
population is generated using a set if tasks and VMs. In climb phase, monkey designated as cloudlet. The mountain is
designated as vm. The monkey will climbs/schedule on the mountain whichever comes first in its path. In Watch and Jump
phase, monkey searches in its local domain to find most optimal solution. This can be related as when a monkey gets on the top
of the mountain, it will check whether other mountains around it higher than its present positions. If present, then jump to that
higher mountain. In somersault phase, each monkey will find a locally maximal mountaintop around its initial point i.e., least
execution cost path is found and monkey rolls down from the mountain. In future better output can be produced using this
algorithm and network bandwidth can also be considered.
6) Bacteria foraging based task scheduling algorithm in cloud environment [15]: For cloud resource scheduling, this algorithm is
an optimization technique based on bacterial foraging. The scheduling parameters used are number of hosts, number of
cloudlets, bandwidth and number of VMs per machine. The main steps are, a virtual machine list is obtained from the data
center after the provisioning of request of the user. Then a random appropriate solution and task lists are initialized. Choosing
the optimal heuristic task is started from the low level. Here each micro organism from the large number of microorganisms
represents an initial solution. Chemotaxis helps microorganisms to construct heuristic steps. At every decision point a fitness
function health is computed. Health will compute and swimming process will start until bacteria have not climbed to long. This
swimming process is continued until the cost has been reduced.
7) Development and Analysis of a three phase cloudlet allocation [16]: Consists of 3 phases:
a) VM categorization phase: In this phase, first select a suitable VM of certain MIPS according to cloudlet length. Then defines the
cloudlet acceptability range for each VM. When cloudlet arrives, DCB selects the VM according to cloudlet’s length.
b) Two round busy checking phase: DCB checks whether the VM is available or not. If available, then check for the balance
threshold condition. If it satisfies the balance conditions, then task is allocated to the VM. If it is not available, two rounds are
performed. Choose a VM with higher MIPS. If VM with higher MIPS present, checks for balance condition and allocates the task
to it. If no, chooses a VM with lower MIPS. Continue the steps.
c) Cloudlet still not allocated phase: If VM is not allocated after the two round busy checking phase, checks for the VM with earliest
finish time. If both VM have same earliest finish time, VM which becomes free earlier is chosen. The disadvantage of this
method is it does not consider the total length of global queue of cloudlets. Lack in finding common completion time among
heterogeneous VMs which may ensure good load balancing, in case of large and small cloudlets is an another problem of this
method.

8) ANGEL: Agent-based scheduling for real-time tasks in virtualized clouds [17]: This method is a novel agent-based scheduling
mechanism in cloud computing environment to allocate real-time tasks and dynamically provision resources. Three phases are
present in the collaborative process i.e., basic matching phase, forward announcement-bidding phase and backward
announcement-bidding phase.
Three agents present:
a) Task Agent
b) VM Agent
c) Manager Agent
Three phases:
i) Basic matching phase: In this basic matching phase when the task arrives, a new task agent is generated. Basic task requirement
information such as task ID, task type, etc. are sent to manager agent from task agent. Task agent is then matched with the VM
agents from VM information board to choose those VMs that satisfy the basic requirements posted by task agents. The matching
process is done by manager agent after receiving the task requirement information. The selected VM information is sent to
corresponding task agents by the manager agent.

©IJRASET: All Rights are Reserved 2264

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue V May 2020- Available at www.ijraset.com

ii) Forward announcement-bidding phase: The manager agent sent the VM’s information to the task agent set. Each task agent
generates forward announcement information. This includes its arrival time, length, deadline, priority, etc. Then send it to
relevant VM agents. The VM agents receive the task’s announcement information .Using this information calculate the
corresponding bidding values based on some rules. The task agents receive the VM agent’s bidding values and make forward
awarding contracts for VM Agents.
iii) Backward announcement-bidding phase: The task agents receive backward announcement from vm agent. This will make
forward contracts with them. The task agents receive the VM agent’s announcement information .Then it will calculate the
corresponding bidding values based on some rules. The VM agents receive the task agent’s bidding values and then it will make
backward awarding contracts. The task is allocated to the VM if a bidirectional contract is built between a VM agent and a task
agent.
Agent-based scheduling algorithm consists of an algorithm for Manager Agent where the manager agent schedules the tasks that
arrive at the same time instant as one batch. Then the manager agent chooses those VMs that satisfy basic requirements of task
agents, and sends the selected VMs’ information to those corresponding task agents. The announcement-bidding phase repeats in
until all the tasks are allocated or rejected.
Another algorithm for task agents is present where each task agent generates forward announcement information to relevant VM
agents. The VM agents return the corresponding bidding values. If it still fails to finish the task before its deadline, the task will be
rejected.
Then an algorithm for VM agent is present. In this algorithm, the VM agents send backward announcement to task agents that send
the forward contract with them, and the task agents return the corresponding bidding values. Then, the VM agents choose a bidder
using the MAX Strategy or the P Strategy, and make backward awarding contracts. By now, a bidirectional contract is built and the
task is allocated to the VM.

9) Dynamic Cloud task Scheduling Based On a two Stage Strategy [18]:

Fig 2: Cloud task scheduling framework based on a two-stage strategy.

In Dynamic cloud task scheduling based on a two stage strategy two stages are present. A Cloud task scheduling framework based
on a two-stage strategy is shown in Fig.2.First stage is to classify tasks based on historical scheduling data .This can be done using a
classifier based on the principle of Byes classifier’s design. On the basis of this, certain numbers of virtual machines of different
types are accordingly created. Time of creating VMs during task scheduling can be saved using this method. Second stage is
matching tasks with concrete VMs dynamically. Dynamic task scheduling algorithms are accordingly proposed to improve
performance of task scheduling. Maintain high level workload balance among all cloud resources in dynamic cloud environment.
Dynamic Cloud Task Scheduling Based on a Two-Stage Strategy will schedules tasks at the idle time slot of VMs as early as
possible. All constraints are also taken into consideration. For tasks of different types, this method has its obvious advantage. When
compared to other methods, when the number and scale of tasks continue to increase, this method increases makespan and average
waiting time more slowly. Using this method, failure rate is lower. Workload of VMs of each VM type as obtained by this method is
better balanced than that by its peers. This method offers higher guarantee ratios of priority and ordinary tasks than its peers do.

©IJRASET: All Rights are Reserved 2265

 International Journal for Research in Applied Science & Engineering Technology (IJRASET)
ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.429
Volume 8 Issue V May 2020- Available at www.ijraset.com

IV. CONCLUSION
In cloud computing environment, resources with different characteristics are served virtually. Efficient scheduling is very important
to manage this type of heterogeneous resources in an optimized way. Various scheduling algorithms and methods of scheduling
have been presented in this paper. This paper deals with some traditional scheduling algorithms and its modified forms. Some
enhanced scheduling methods are also discussed in this paper. This comparative study will be helpful in selection of appropriate
scheduling algorithms for different cloud consumers as well as cloud service providers. To enhance the overall efficiency of cloud
computing environments in future, understanding various scheduling methods may useful for development of other more efficient
algorithms.

REFERENCES
[1] Ruba Abu Khurma, Heba Harahsheh and Ahmad Abdel-Aziz Sharieh, “Task Scheduling Algorithm in Cloud Computing Based on Modified Round Robin
Algorithm,” Journal of Theoretical and Applied Information Technology ,, 2018.
[2] S.Devipriya and C.Ramesh ,“Improved Max-Min Heuristic Model For Task Scheduling in Cloud”, International Conference on Green Computing,
Communication and Conservation of Energy (ICGCE),, 2013.
[3] HE XiaoShan, SUN XianHe and Gregor yon Laszewski,“QoS Guided Min-Min Heuristic for Grid Task Scheduling”, Journal of Computer Science and
Technology, 2003.
[4] Kobra Etminani and Prof. M. Naghibzadeh, “A Min-Min Max-Min Selective Algorithm for Grid Task Scheduing”, Third IEEE/IFIP International Conference
on Internet, Uzbekistan, 2007.
[5] O.M.Elzeki, M.Z.Reshad and M.A.Elsoud, “Improved Max-Min Algorithm in Cloud Computing”, International Journal of Computer Applications, July, 2012.
[6] Upendra Bhoi1and Purvi N. Ramanuj, “Enhanced Max-Min Task Scheduling Algorithm in Cloud Computing,”, International Journal of Application or
Innovation in Engineering Management, 2013.
[7] Saeed Parsa and Reza Entezari-Maleki, “RASA:A New Grid Task Scheduling Algorithm,”, International Journal of Digital Content Technology and its
Applications, 2009.
[8] Sunilkumar Nakum, C. Ramakrishna and Amit Lathigara, “Reliable RASA Scheduling Algorithm for Grid Environment,”,2014 IEEE International Conference
on Computer Communication and Systems, 2014.
[9] Yang Cui and Zhang Xiaoqing, “ Workflow Tasks Scheduling Optimization Based on Genetic Algorithm in Clouds “,, 3rd IEEE International Conference on
Cloud Computing and Big Data Analysis, 2018.
[10] Airo Alonso Giraldo and M. Passino, “Honey Bee Social Foraging Algorithm for Resource Allocation”, Springer Handbook of Computational Intelligence,
2015.
[11] H.W.Hashem and R. Rizk, “Honey Bee Based Load Balancing in Cloud Computing,”, Transaction on Internet Information System, 2017.
[12] Jaspinder Kaur and Brahmaleen Kaur Sidhu, “A New Flower Pollination based Task Scheduling Algorithm in Cloud Environment”, 4th International
Conference On Signal Processing Computing And Control, 2017.
[13] Shridhar Domanal, Ram Mohana Reddy Guddeti and Rajkumar Buyya, “A A Hybrid Bio-Inspired Algorithm for Scheduling and Resource Management in
Cloud Environment”, IEEE TRANSACTIONS ON SERVICES COMPUTING, 2016.
[14] Punit Gupta and Prateek Tewari, “Monkey Search Algorithm for Task Scheduling in Cloud IaaS”,Fourth International Conference on Image Information
Processing, 2017.
[15] Juhi Verma, Srichandan Sobhanayak and Suraj Sharma,“Bacteria Foraging Based Task Scheduling Algorithm in Cloud Computing Environment”, International
Conference on Communication and Automation, 2017.
[16] Sudip Roy,Sourav Banerjee,K.R.Chowdhury and Utpal Biswas, “Development and Analysis of a Three Phase Cloudlet Allocation Algorithm,”, Journal of King
Saud University - Computer and Information Sciences, 2017.
[17] Xiaomin Zhu, Chao Chen, Laurence T. Yang and Yang Xiang, “ANGEL: Agent-Based Scheduling for Real-Time Tasks in Virtualized Clouds”, IEEE
Transactions on Computers, 2015.
[18] PeiYun Zhang and and MengChu Zhou, “Dynamic Cloud Task Scheduling Based on a Two-Stage Strategy,” IEEE TRANSACTIONS ON AUTOMATION
SCIENCE AND ENGINEERING, 2018.

©IJRASET: All Rights are Reserved 2266