Received March 9, 2022, accepted March 24, 2022, date of publication March 30, 2022, date of current version

April 7, 2022.
Digital Object Identifier 10.1109/ACCESS.2022.3163273

Multiobjective Task Scheduling in Cloud

Environment Using Decision Tree Algorithm
HADEER MAHMOUD 1,2 , MOSTAFA THABET3 , MOHAMED H. KHAFAGY1 ,
AND FATMA A. OMARA4 , (Member, IEEE)
1 Department of Computer Science, Faculty of Computers and Information, Fayoum University, Fayoum 63514, Egypt
2 Facultyof Information Systems and Computer Science, October 6 University, Giza 12585, Egypt
3 Department of Information System, Faculty of Computers and Information, Fayoum University, Fayoum 63514, Egypt
4 Department of Computer Science, Faculty of Computers and Artificial Intelligence, Cairo University, Cairo 12613, Egypt

Corresponding author: Hadeer Mahmoud ([email protected])

ABSTRACT In recent years, Cloud computing has been developed and become the foundation of a wide
range of applications. It allows users to access a catalog of standardized services and respond to their business
needs flexibly and adaptively, in the event of unforeseen demands, paying solely for the consumption they
have made. Task scheduling problem is considered one of the most critical cloud computing challenges.
The problem refers to how to reasonably order and allocate the applications tasks provided by the users
to be executed on virtual machines. Furthermore, the quality of scheduling performance has a direct
effect on customer satisfaction. The task scheduling problem in cloud computing must be more accurately
described in order to improve scheduling performance. In this paper, a multi-objective task scheduling
algorithm is proposed based on the decision tree in a heterogenous environment. We introduce a new Task
Scheduling-Decision Tree (TS-DT) algorithm for allocating and executing an application’s task. To evaluate
the performance of the proposed TS-DT algorithm, a comparative study was conducted among the existing
algorithms; Heterogeneous Earliest Finish Time (HEFT), Technique for Order of Preference by Similarity to
Ideal Solution that incorporates the Entropy Weight Method (TOPSIS-EWM), and combining Q-Learning
with the Heterogeneous Earliest Finish Time (QL-HEFT). Our results show that the proposed TS-DT
algorithm outperforms the existing HEFT, TOPSIS-EWM, and QL-HEFT algorithms by reducing makespan
by 5.21%, 2.54%, and 3.32%, respectively, improving resource utilization by 4.69%, 6.81%, and 8.27%,
respectively, and improving load balancing by 33.36%, 19.69%, and 59.06%, respectively in average.

INDEX TERMS Cloud computing, task scheduling, data dependency, decision tree, makespan, resource
utilization, load balancing, energy consumption.

I. INTRODUCTION resource efficiency. Task scheduling depends on the existence

Cloud computing is a modern computer technology that of dependencies among the tasks. The problem of scheduling
employs virtualized infrastructure to provide secure and reli- dependent tasks in a heterogeneous environment has drawn
able services to end-users in a complicated environment. many attention of researchers in this area. Among the most
Because it provides essential information technology (IT) studied area is the Directed Acyclic Graph (DAG), which is
services such as computing resources in the form of virtual the most common graph that shows the dependability of the
machines (VMs), cloud computing has gotten a lot of atten- application’s tasks. This dependent task scheduling is also
tion as a computing model [1], [2]. called DAG scheduling.
Unfortunately, Cloud computing has some challenges such Some of the DAG types are Montage, the Srna Identifi-
as performance, resource management, cost, etc. [3]. On the cation Protocol using High-throughput Technology (SIPHT),
other hand, task scheduling on cloud computing is the allo- Cyber-Shake, Epigenomics, and the Laser Interferometer
cation of users’ tasks on the available resources to optimize Gravitational Wave Observatory (LIGO) workflow applica-
the execution time, enhance load balancing, and increase tions [4]–[6]. (See Figure 1).
In DAG scheduling, the workflow is presented by
The associate editor coordinating the review of this manuscript and G = (T), where T = {t1 , t2 . . . ,} is the set of tasks and
approving it for publication was Muhammad Zakarya . E = {e1, e2 . . . ,} is the set of edges. Each task ti ∈ T denotes

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
36140 VOLUME 10, 2022
H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

distributed and parallel computing environments, a set of

scheduling strategies has been proposed by many researchers.
K. Naik et al. [11] have described a new hybrid multi-
objective heuristic method that combines Non-dominated
Sorting Genetic Algorithm-II (NSGA-II) and Gravitational
Search Algorithm (GSA) called as NSGA-II & GSA to assist
with VM selection for application scheduling. NSGA-II has
the ability to expand the search space through exploration,
FIGURE 1. Structure of DAG workflow [7]. while GSA has the ability to exploit the good solution to
discover the best solution and so avoid the algorithm getting
trapped in local optima. This hybrid algorithm is designed
an application task, and each E (i, j) ∈ E represents the to achieve the fastest response time and lowest cost for
communication cost between the independent tasks, where scheduling a larger number of tasks with the minimum total
task ti is should be executed before task tj . A structure of the energy consumption. Unfortunately, there is no load balanc-
DAG of some workflow types is shown in Figure 1. ing between VMs.
Scheduling and allocation of the applications’ tasks are S.Pang and colleagues [12] have developed a hybrid
considered Non-deterministic Polynomial (NP)-Complete scheduling algorithm based on the Estimation of Distribu-
problems [8]. Therefore, optimization approaches are used to tion Algorithm (EDA) and Genetic Algorithm (GA). The
solve these problems by considering performance parameters algorithm initially generates some feasible solutions using
such as makespan, load balance, resource usage, cost, power EDA operations, then utilizes GA operations to generate new
consumption, etc., to solve these problems [9]. solutions based on the great solutions selected in the previous
Machine learning is currently being used in various fields phase to expand the search range of solutions, and finally
such as speech recognition, data classification, and face selects the best solution. The purpose of this technique is to
recognition [10]. It has played a significant role in task reduce the task completion time and enhance load balancing.
scheduling and several other areas in computer science tech- However, this paper does not consider the dynamics and
nology. Among the popular machine learning techniques used uncertainties of the cloud computing environment.
for developing and visualizing predictive models and algo- S.H.H. Madni et al. [13] have presented a novel Multi-
rithms is the decision tree. objective Cuckoo Search Optimization (MOCSO) technique
In this paper, a Task Scheduling-Decision Tree (TS-DT) for dealing with the cloud computing resource scheduling
algorithm, which is a task scheduling algorithm based on the problem. The goal of this technique is to explore the multi-
decision tree is introduced. The performance of the proposed objective resource scheduling problem in an Infrastructure as
task scheduling is evaluated using load balance parameters a service (IaaS) cloud computing environment by maximizing
such as makespan, resource utilization, and power consump- resource consumption. The load balancing across VMs is
tion in a heterogeneous environment. By the work in this considered a big flaw in this technique.
paper, the following significant contributions are satisfied: Y.Q. Han and Q. Li. Jun [14] have solved the flexible task
scheduling problem in a cloud system by proposing a hybrid
• A new task scheduling algorithm based on the decision
discrete Artificial Bee Colony (ABC) algorithm. The sug-
tree method for scientific workflows is proposed.
gested algorithm includes three categories of artificial bees;
• Extensive simulation tests on various performance met-
employed, onlooker, and scout bees, as in the classical ABC
rics are used to evaluate the proposed algorithm.
algorithm. The proposed ABC algorithm reduces completion
The rest of this paper is organized as follows; Section 2 dis- time and improves balancing machine loads. One of this
cusses the related works. In Section 3, the principles of algorithm’s major flaws is resource utilization.
the proposed algorithm are described. Section 4 illustrates Avinash Kaur et al. [15] have proposed a new workflow
the proposed algorithm in detail. The CloudSim simulator’s scheduling scheme by integrating the Deep Q-learning mech-
configuration and the proposed Decision Tree Algorithm are anism and the HEFT algorithm is called DQ-HEFT. The
discussed in Section 5. The comparative study among the scheme is considered the most common heuristic scheduling
proposed algorithm and other existing algorithms like HEFT, in literature. The algorithm consists of two phases: gaining
TOPSIS-EWM, and QL-HEFT are discussed in Section 6. the task’s execution order at each stage and allocating the task
Finally, the conclusion and future work are given to the processor with a significantly higher volume of data.
in Section 7. It is worthy to note that the DQ-HEFT algorithm can achieve
better makespan and speed. However, the main drawback of
II. RELATED WORK the DQ-HEFT algorithm is that excessive value updates of
Task scheduling in distributed, parallel, and cloud comput- the Q-table are performed in large-scale task optimization
ing environments have become an important research topic. problems, which slow down the scheduling process.
Its purpose is to ensure an effective distribution of com- A. Al-maamari and F. Omara [16] have proposed a task
puting resources to provide high performance. In traditional scheduling algorithm for the cloud computing environment.

VOLUME 10, 2022 36141

 H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

The algorithm is considered a fusion of the Cuckoo The first step is used to calculate the VM’s capacity and load
Search (CS) algorithm and the Dynamic PSO (DAPSO) (under-full VM, balanced VM, high-balance VM, and over-
algorithm, which has been modified to increase the popu- loaded VM). In the second step, for each VM, the required
lation. According to this algorithm, tasks are assigned to time is determined to execute the task. In the third step, based
virtual machines (VMs) to minimize makespan and maxi- on the VM state and the task time, a decision is made to spread
mize resource uses. However, there is no load unbalancing tasks. Unfortunately, resource utilization between VMs is
between VMs. considered a critical weakness of this algorithm.
Arabi Keshk et al. [17] have introduced Modified Ant Zeshan Iqbal et al. [22] have proposed an algorithm called
Colony Optimization for Load Balancing (MACOLB) algo- Parental Prioritization Earliest Finish Time (PPEFT) for a
rithm to allocate the incoming jobs to the virtual machines heterogeneous distributed environment. The algorithm con-
(VMs). The tasks are allocated to the VMs based on the sists of two phases; the prioritization of the tasks and the
processing powers (i.e., tasks are allocated in descending processor’s assignation. First, the tasks are scheduled in
order, starting from the most powerful VM, and so on) by the Parental Priority Queue (PPQ) based on the descending
considering balancing VMs’ loads. The MACOLB is used to Rank and parental priority in the task prioritization phase.
find the proper resource allocation for batch tasks in the cloud Then, the Processor Assigning Phase assigns each task in the
system, minimize the makespan, and achieve better system PPQ queue to a processor that guarantees fast execution
load balance. However, the resource utilization between VMs (i.e., minimum computation cost). Experimentally, the
is considered a crucial flaw of this algorithm. PPEFT scheduling algorithm performs substantially better
Jae-Min Yu1 and colleagues [18] have proposed a decision concerning cost and schedule makespan than other algo-
tree-based method for scheduling flexible workshops with rithms. Unfortunately, load balancing between VMs is a crit-
multiple process plans. For static and dynamic flexible job ical weakness of this algorithm.
shops, two decision tree-based scheduling mechanisms were S.C. Sharma et al. [23] have modified the HEFT algo-
created. All jobs were provided in advance in the static case, rithm to effectively distribute the workload between proces-
and the decision tree is used to select a priority dispatching sors and effectively reduce completion time. This algorithm
rule to process all of them. In the dynamic scenario, jobs analyzes various algorithms for the task scheduling, param-
arrive over time. The decision tree is used to select a priority eters, tools, improvement, and algorithm limitations. This
rule in real-time according to a rescheduling strategy using algorithm reduces makespan and improves load balancing by
a decision tree that is modified regularly. The objectives comparing it to the existing HEFT and the Critical Path on
considered in this method are makespan, total flow time, and a Processor Algorithm (CPOP) [24] algorithms. The critical
total delay, but the load balancing between VMs was not weakness of this algorithm is the sleek time.
considered. In this paper, the decision tree has been used to optimize
Liu Yuan, Dong Yinggang, etc. [19] have proposed a the multi-objective task scheduling problem by minimizing
static HEFT task scheduling algorithm, called ST-HEFT. The makespan, satisfying load balancing among virtual machines,
algorithm consists of two key steps; task sorting and task and maximizing resource utilization.
mapping. According to the sorting step, tasks are sorted based
on the maximum communication cost between them and their III. OVERVIEW
direct VMs. The task mapping step is assigned to the VM that In this paper, we proposed a task scheduling algorithm based
provides the earliest execution time. The proposed algorithm on the decision tree for a heterogeneous cloud environment
has achieved better performance by reducing the development has been proposed. Also, we evaluated the performance of the
threshold for parallel computing programs and increasing the proposed algorithm through a comparative study among the
utilization of various computing devices’ capabilities in the HEFT, TOPSIS-EWM, and QL-HEFT algorithms, which are
heterogeneous computing environment. On the other hand, widely used algorithms for task scheduling in the cloud com-
load balancing and sleek time are the critical weaknesses of puting environment. Based on the above-mentioned goals,
this algorithm. we now discuss the principles of the existing algorithms in
Sambit Kumar Mishra et al. [20] have suggested an the following sections.
Adaptive Task Allocation (ATAA) algorithm in the cloud
environment. This algorithm uses the Expected Time to Com- A. THE HETEROGENEOUS EARLIEST FINISH TIME
pletion (ETC) matrix to solve the heterogeneous environment ALGORITHM
problem, including completing all tasks on VMs. The author In the HEFT algorithm [25], the tasks presented in the
uses a technique that reduces energy consumption and mini- DAG are scheduled to a series of heterogeneous machines.
mizing the makespan of the system. Also, the major weakness This algorithm consists of two phases; ranking and proces-
of this algorithm is the load balancing between VMs. sor selection phases. The goal of the ranking phase is to
Atyaf Dhari et al. [21] have proposed a cloud comput- provide a priority for each task. The Processor Selection
ing environment load balancing decision algorithm, called phase concerns about allocating each task to a suitable pro-
(LBDA), to enhance load balancing among virtual machines cessor. This phase will be repeated until all tasks will be
and reduce makespan. The algorithm consists of three steps. scheduled for the available processors [26]. In the ranking

36142 VOLUME 10, 2022

H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

phase, the upward (ranking) function is used to define the other hand, the decision tree is a rooted tree having leaf
priority of each task which is defined recursively by using and non-leaf nodes. The decision criteria for classification
Equation (1) [27]: and regression trees distinctly depend on the decision tree.
Meanwhile, a decision tree is a rooted tree with leaf and
Rank (t i ) = wl + maxtj ∈succ(ti ) cl,J + Rank t j



(1)
non-leaf nodes. The leaf nodes represent the classification or
where Wi is the average of the computation cost of the task decision-making, whereas the non-leaf nodes represent the
ti , ci,j is the average of the communication cost between the selection options by dividing the instance space into two or
edges from ei to ej , and succ ( ti ) is the set of successors of the more subspecies based on a discrete function of input attribute
task ti . It’s important to remember that Rank (ti ) is determined values (See Figure 2) [31].
by the computation of all its children’s Rank (tj ).
In the ‘‘processor selection’’ phase, tasks are sorted in
descending order according to their rank values. Then, the
processors assign tasks by selecting the processor with the
shortest finish time for each task. However, the HEFT algo-
rithm always considers the processor with the earliest finish
time to allocate tasks, but it does not consider load balancing
among the processors [28].

B. A MULTI-CRITERIA DECISION-MAKING APPROACH

A Technique for Order of Preference by Similarity
FIGURE 2. Decision tree structure [32].
to Ideal Solution (TOPSIS) workflow scheduling algo-
rithm in a cloud environment has been introduced that The decision tree is a popular method for creating and
incorporates the Entropy Weight Method (EWM) called visualizing predictive models and algorithms [23], [33].
(TOPSIS-EWM) [13]. The proposed algorithm aims to As explained earlier, the static approach uses the decision
reduce makespan, cost, and energy consumption while tree to select a priority rule combination to process the set
increasing reliability. According to the ROPSIS-EWM algo- of given tasks, i.e., no rule changes over the scheduling
rithm, EWM is used to determine the input weight of the horizon. Hence, it can be used for planning purposes. The
attributes schedule length (EFT), cost, reliability, and energy static decision tree-based mechanism suggested in this study
consumption. The TOPSIS approach is then used to choose is shown in the next section.
the optimal virtual machine for each task. The research takes
into consideration a cloud environment with dynamic voltage IV. PROBLEM STATEMENT
scaling (DVS) and pay-per-use heterogeneous VM instances. Multi-objective optimization has a significant impact on
The MIPS of the VM instances is used in the simulations scheduling issues in the cloud computing environment.
directly proportional to VM pricing. The (TOPSIS-EWM) We covered three relatively approaches for resolving task
algorithm does not consider any user preferences, such as scheduling performance, each has its own set of restrictions:
deadlines, load balance, and resource utilization. • When allocating tasks, the HEFT method always con-
siders the processor with the earliest finish time, but it
C. QL-HEFT ALGORITHM ignores load balancing among the processors.
Authors in [29], proposed a novel task scheduling algorithm • The (TOPSIS-EWM) algorithm considers the input
that reduces the makespan by combining Q-Learning with the weight of the attributes schedule length like (EFT), cost,
HEFT algorithm is called (QL-HEFT). The algorithm uses reliability, and energy consumption. Then, the TOPSIS
the upward rank value of HEFT as the immediate reward. technique is used to select the best virtual machine for
In the Q-learning framework, the agent can obtain better each task without considering any user preferences like
learning results through the self-learning process to update deadlines, load balance, and resource utilization.
the Q-table. The QL-HEFT algorithm is divided into two • The QL-HEFT approach for solving large-scale task
phases: the task sorting phase and the processor allocation optimization issues has some limitations, such as a big
phase. The task sorting phase uses Q-learning to find the Q-table that causes long update times that effect on the
best order of the tasks, while the processor allocation phase Makespan in the task schedule.
uses the earliest finish time strategy. However, the authors
discovered that using the QL-HEFT algorithm to solve large- V. THE PROPOSED TASK SCHEDULING DECISION TREE
scale task optimization problems has some drawbacks, such (TS-DT) ALGORITHM
as an excessively large Q-table that causes long update times. In this section, we introduce the TS-DT algorithm to reduce
the makespan, enhance load balance, and maximize utiliza-
D. A DECISION TREE DEFINITION tion of the resource. The algorithm consists of three phases:
A decision tree is a hierarchical data structure that uses a the priority task, the resource matrix, and the resource allo-
divide-and-conquer technique to represent data [30]. On the cation phase. First, the task priority phase is used to assign

VOLUME 10, 2022 36143

 H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

a rank for each task. The resource matrix phase is used in Feature 3: Total length of tasks (TTL) assigned to each
collecting the tasks’ features in the form of a matrix, while VM. This refers to the length of the instructions
the resource allocation phase is where tasks are scheduled on of tasks on VM.
the proper VMs using the decision tree. The principles of each Feature 4: Showing the VM-based task parent (TP) (i.e.,
phase will be explained in the following sections. Parent location (0/1)), where one is typed if there
is a parent for the task; otherwise, zero is typed.
A. TASK PRIORITY PHASE This feature considers the communication cost
According to the task priority phase, a rank is assigned for between tasks.
each task in the given workflow (i.e., DAG). By considering For example, a structure of a task structure with its features
tasks set T = {T1, T2, . . . , Tn}. If Ti < Tn , then Ti is the is summarized in a matrix (T) by considering five VMs with
parent of Tn . Equation (1) has been changed by adding task four features, as shown in Figure 3.
length (TL ), which indicates the length of the instruction of
a cloudlet (i.e., task) to be processed in the virtual machine
(VM), and the number of child’s (Nc ) (See Equation (2)).
Therefore, the Rank of each task in the given workflow is
defined using Equation (2).
FIGURE 3. A Structure of a tasks’ matrix.
Rank (ti ) = wl + maxtj ∈succ(ti ) cl,j + Rank tj



+ Tl + Nc
(2)
C. RESOURCE ALLOCATION PHASE
After assigning priority to each task, the tasks will be sorted In this phase, the proper VMs are selected to execute tasks
in descending order according to their Rank value and stored in the Rank [T] list, which contains the tasks which order in
in the Rank [T] list. As a result, the most important task will descending order according to their priorities.
be executed first. The pseudo-code of the Task Priority phase For the task that plays a role in the Rank [T] list (i.e., the
is as follows: task with high priority), the decision tree is constructed to
represent its features from its task matrix. In the case of leaf
TABLE 1. Shows the pseudo-code of the tasks priority phase. nodes, a test is done to check if the task’s parent is on the same
VM or not according to Feature 4. If the answer is ‘‘yes’’, the
communication cost remains zero. In the case of ‘‘no’’, the
communication cost is considered between the parent and
the successor.
The output in the leaf nodes is the summation of the task’s
features in the task’s matrix, which is defined using Feature 1
to Feature 3 (i.e., CP, EFT, and TTL) with considered the
summation of the communication cost from the DAG work-
flow if the task’s parent is not on the same VM. If the parent
task in the same VM, it is compute using Equation (3). When
the parent is not in the same VM, it is calculated using
Equation (4).
VM = CPi + EFT + TTL (3)
VM = CC(i,j) + CPi + EFT + TTL (4)
where CC(i,j ) is the communication cost between ti that
presented in Rank [T] list with tj is the last task in VM. (CPi )
is the computation cost for the task ti among all the VMs.
(EFT) is the earliest finish time of all tasks in VM. (TTL) is
B. RESOURCE MATRIX PHASE the total length of tasks assigned to each VM.
This phase is used to collect the features of the selected Finally, the task is assigned to the VM, which has the
task from the Rank [T] list and store them in the task’s lowest value that will come out of the tree’s leaf nodes
matrix (T). In the T matrix, columns represent the number (See Figure 4). The pseudo-code for the Resource Allocation
of needed resources, while rows represent four features for phase is shown in Table 2.
each task as follows: Example: To explain how the proposed TS-DT algorithm
Feature 1: Computation cost (CP) of each task on each VM. works, a sample of the task graph and the computational
Feature 2: Assigning the task to VM by selecting VM with task costs on each VM with considering 3 VMs are depicted
Earliest Finish Time (EFT). in Figure (5a, b).

36144 VOLUME 10, 2022

H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

TABLE 3. Calculation Rank [T] for Figure 5.

FIGURE 4. The proposed TS-DT algorithm using decision tree.

TABLE 2. Shows the pseudo-code of the resource allocation phase.

FIGURE 6. The resource features of task T1 .

3) PHASE 3: RESOURCE ALLOCATION PHASE

In this phase, each task is assigned to a suitable VM based on
its decision tree. The cost of communication value between
tasks and their parent and successor will be assessed based
on the matrix that constructs in Phase 2. According to the
decision tree of task (T1), the summation value of VM3 is
considered the lowest value (i.e., 0+10=10) (See Figure 7).
So, VM 3 is assigned to execute the task (T1).

FIGURE 7. The decision tree of task T1 .

Phases two and three (i.e., Resource Matrix Phase and

FIGURE 5. A task graph and computational cost example. Resource Allocation Phase) are repeated for each task in the
Rank[T] list until all tasks are assigned to VMs.
1) PHASE 1: TASK PRIORITY PHASE As a result, the entire workflow makespan is
By applying this phase, a rank is assigned to each task using 413 milliseconds in both the task priority and resource alloca-
Equation (2), and the tasks are sorted in descending order in tion phases. The deviation of the load balance is zero because
Rank [T] (See Table 3). the term of the Resource Matrix phase considers the number
of tasks on each VM. The resource utilization rate is 94.67 %,
2) PHASE 2: RESOURCE MATRIX PHASE and it appears that the Resource Matrix phase works better.
In this phase, the features of the selected task from the (See Figure 8).
Rank [T] list will be collected and stored in the task’s matrix
by considering T1 in Figure 5, Feature2 (i.e., EFT), Feature3 VI. THE PERFORMANCE EVALUATION
(i.e., TTL), and Feature4 (i.e., TP) are considered zero on In this section, we discuss the performance matrices, exper-
all VMs because it is the entry task. Therefore, the features imental environment, and the benchmark used in evaluating
of (T1 ) in its matrix are presented in Figure 6. the performance of the proposed TS-TD algorithm.

VOLUME 10, 2022 36145

 H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

6) TOTAL EXECUTION TIME (TET)

It represents the summation between the busy time and idle
time for each VMi. It is defined using Equation (10) [39].
TETvm = Tbusy (VMi ) + Tidle (VMi ) (10)

7) POWER CONSUMPTION
The power consumption of VMi consists of two parts; busy
power consumption and idle power consumption. It is calcu-
lated using Equation (11) [40].
FIGURE 8. Using TS-DT algorithm, the makespan is 413 msec.
Xn  

PC VMj = Pbusyj ∗ Tbusy + Pidlej ∗ Tidle (11)
0

where Pbusyj is the busy power of the machines with 220W,

A. PERFORMANCE METRICS Pidlej is the idle power of the machines with 95W, Tbusy is
The performance matrices which are used to measure the the busy time of the machine, and Tidle is the idle time of the
efficiency of the task scheduling algorithms on the cloud machine.
computing environment includes the following:
8) IMPROVEMENT RATE (IRx )
1) MAKESPAN Makespan, resource utilization, and load balance are factors
It displays the maximum completion time of the schedule. (x) that will be considered to determine the performance
This parameter is defined by Equation (5) [34]. improvement of the proposed TS-DT algorithm relative to the
current HEFT, TOPSIS-EWM, and QL-HEFT. We calculate
Makespan = max (Cti )ti T (5)
IRx using Equation (12) [41].
where Cti is the execution time of the longest task ti , and T IRx (%)
is the number of the tasks on the workflow of an application. 
x(existing algorithm) − x(proposed algorithm

=−
x(existing algorithm)
2) RESOURCE UTILIZATION RATE (RU)
∗100 (12)
We define RU as the ratio of the total consumed time by VMs
to the makespan of the parallel application. It is calculated as
9) AVERAGE DEVIATION
a percentage using Equation (6) [35].
P The ratio between the total summation of IRx for each VMi
VMi BusyTime over their number to get the average deviation from the ideal
RU (VMi ) % = ∗ 100 (6)
Makespan rate in percentage is calculated using Equation (13) [42].
Xn  IRx (VMi ) 
3) LOAD BALANCING (LB) Average(%) = (13)
This is the ratio of the total number of tasks to the number of
i=1 n
VMs. We calculate LB using Equation (7) [36].
B. THE EXPERIMENTAL ENVIRONMENT
Number of Tasks The CloudSim 3.0.2 toolkit is an open-source simulator
LB = (7)
Number of VMs developed by WorkflowSim 1.0 on Windows 7 OS with a
Core i7 2.70 GHz processor [43]. The CloudSim simulator is
4) BUSY TIME used to implement and evaluate the performance of the pro-
It represents the task length indicates the instructions length posed TS-DT Algorithm. The Eclipse IDE 4.12.0 is used to
of a cloudlet to be processed on each VMi . It is defined using run CloudSim 3.0.2. The used benchmarks are Montage_25,
Equation (8) [37]. SIPHT_30, Cyber- Shake_30, and Epigenomics_24
P
tlength workflows [44].
Tbusy (VMi ) = (8)
V Mi
P VII. PERFORMANCE ANALYSIS
where length is the summation of each task length. To evaluate the performance of the proposed TS-DT algo-
rithm, a comparative study was conducted among HEFT,
5) IDLE TIME TOPSIS-EWM, QL-HEFT algorithms, and the proposed
This is the difference between the total execution time TS-DT algorithm in terms of makespan, load balance,
and the busy time for each VMi . It is determined using resource utilization, and power consumption. The tasks are
Equation (9) [38]. considered dependent, and each task has different charac-
teristics, such as length, id, start time, and finish time. The
Tidle (VMi ) = Ttotal − Tbusy (9) implementation was carried out with the consideration of 5,

36146 VOLUME 10, 2022

H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

10, 20, and 40 VMs using Montage_25, SIPHT_30, Cyber-

Shake_30, and Epigenomics_24 workflows. We focus on the
hypervolume indicator to measure the quality of a set of
trade-off solutions [45].

A. MAKESPAN EVALUATION
The implementation results of the proposed TS-DT, HEFT,
TOPSIS-EWM, and QL-HEFT algorithms with respect to
makespan using Montage_25, SIPHT_30, Cyber-Shake_30,
and Epigenomics_24 workflows by 5, 10, 20, and 40 VMs
using Equation (5), as shown in Figure 9 from (a) to (d).
According to the implementation results in Figure 9, it is
found that the proposed TS-DT algorithm outperforms the
HEFT, TOPSIS-EWM, and QL-HEFT algorithms. This is
because of the following reasons:
• During the Task Priority phase, the proposed TS-DT
algorithm determines the proper task to be executed by
increasing its priority while using the task length and
number of childs.
• During the Resource Allocation phase, the decision tree
and the summation of the features in the task’s matrix
are used to select the VM with the lowest value.
• Some features are used to enhance the makespan
(i.e., computation cost, Earliest Finish Time (EFT), and
parent location).
The average improvement rate of makespan, in percent,
of the proposed TS-DT algorithm compared to existing
HEFT, TOPSIS-EWM, and QL-HEFT algorithms were deter-
mined using Equation (12) and presented in Table 4.

TABLE 4. Improved average rate of makespan (in %) of the proposed

TS-DT algorithm.

According to the comparative results in Table 4, We found

that the proposed TS-DT algorithm outperforms, in average,
the default HEFT, TOPSIS-EWM, and QL-HEFT algorithms
by approximately 5.21%, 2.54%, and 3.32%, respectively.

B. RESOURCE UTILIZATION
TS-DT, HEFT, TOPSIS-EWM, and QL-HEFT algo-
rithms with respect to resource utilization using Mon- FIGURE 9. Comparative results for makespan.
tage_25, SIPHT_30, Cyber-Shake_30, and Epigenomics_24
workflows using 5, 10, 20, and 40 VMs are presented in
Figure 10. Based on the comparison results in instruction’s length of tasks on VM considers the devices’
Figure 10 (a)-(d) (See Equation (6)), we confirmed that the consumption rate.
proposed TS-DT algorithm outperforms the existing algo- The average improvement of the proposed TS-DT algo-
rithms in terms of resource utilization for any number of rithm in terms of resource utilization in percentage in relation
VMs. This is possible because, during the Resource Matrix to the existing HEFT, TOPSIS-EWM, and QL-HEFT algo-
phase, the proposed TS-DT algorithm selects the minimum rithms were determined using Equation (12) and presented
of the total length of tasks assigned to each VM. The total in Table 5.

VOLUME 10, 2022 36147

 H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

TABLE 5. Improved average rate of resource utilization (in %) for the

proposed TS-DT algorithm.

SIPHT_30, Cyber-Shake_30, and Epigenomics_24 work-

flows using 5, 10, 20, and 40 VMs are presented in Figure 11
from (a) to (d).
According to the comparison results in Figure 11 is found
that the proposed TS-DT algorithm outperforms the existing
other algorithms in terms of load balancing. This is because
that during the Resource Matrix phase (Feature 3), the pro-
posed TS-DT algorithm considers the length of the total task
on each VM when assigning a new task to VM. Finally, the
TS-DT select the minimum value. So, the phase is influenced
by the device load at similar rates.
The average load balance improvement in the percentage
of the proposed TS-DT algorithm relative to the existing
HEFT TOPSIS-EWM, and QL-HEFT algorithms has been
determined using Equation (13) and illustrated in Table 6.

TABLE 6. Improved average rate of load balance (in %) of TS-DT

algorithm.

According to the comparative results in Table 6, it is found

that the proposed TS-DT algorithm outperforms, in average,
the default HEFT, TOPSIS-EWM, and QL-HEFT algorithms
by approximately 33.36%, 19.69%, and 59.06%, respectively.

D. POWER CONSUMPTION
In the heterogeneous cloud platform, the power consump-
tion consists of two components; busy power consumption
FIGURE 10. Comparative results for resource utilization. and idle power consumption. During the power process-
ing operations, the busy power consumption is the energy
According to our comparative results in Table 5 show that consumed, and the idle power consumption is the energy
the proposed TS-DT algorithm outperforms, in average, the consumed during the VM in the idle state [40]. In order
default HEFT, TOPSIS-EWM, and QL-HEFT algorithms by to accurately calculate the power consumption, the work
approximately 4.69%, 6.81%, and 8.27%, respectively. in this paper estimates the busy power consumption and
idle power consumption of the VM, respectively. Then,
C. LOAD BALANCING the total power consumption of the VM is obtained using
The load balancing of each VM is calculated by the ratio Equations (8, 9, 10, 11, 12, and 13). According to the imple-
of the number of tasks and the total execution time in VMs mentation results, it is found that our proposed TS-DT algo-
(See Equation (7)). The implementation results of our pro- rithm increases power consumption relative to the HEFT,
posed TS-DT, HEFT, TOPSIS-EWM, and QL-HEFT algo- TOPSIS-EWM, and QL-HEFT algorithms, this is because the
rithms with respect to load balancing using Montage_25, following reason:

36148 VOLUME 10, 2022

H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

TABLE 7. Average in power consumption (in %) of the proposed TS-DT

algorithm.

FIGURE 12. The hypervolume of the compared algorithms for Montage.

FIGURE 13. The hypervolume of the compared algorithms for

Epigenomics.

VMs specifications, which increases the power con-

sumption rate of the devices.
According to the implementation results in Table 7, it is
found that the proposed TS-DT algorithm increases the
power consumption relative to the default HEFT, TOPSIS-
EWM, and QL-HEFT algorithms by approximately 12.89 %,
43.52 %, and 28.38 %, respectively. This is considered the
main limitation of the proposed TS-DT.
Generally, the average Makespan, resource utilization, load
balance, and power consumption of the proposed TS-DT,
HEFT, TOPSIS-EWM, and QL-HEFT algorithms are shown
in Tables (4,5, 6, and 7).
FIGURE 11. Comparative results for load balance.
With respect to makespan, resource utilization, and load
balance, the performance of the TS-DT algorithm is better
• The proposed TS-DT algorithm selects the minimum than that of HEFT, TOPSIS-EWM, and QL-HEFT algorithms
of CP and VM with EFT for the assigned task to VM in most cases, which is also confirmed by the hypervol-
during the Resource Matrix phase. This means that ume as shown in Figures (12, 13) for Montage and Epige-
the proposed algorithm consumes the high available nomics workflow. This is because the HEFT algorithm is a

VOLUME 10, 2022 36149

 H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

single-objective scheduling algorithm and does not consider [9] M. Lavanya, B. Shanthi, and S. Saravanan, ‘‘Multi objective task schedul-
other objectives such as load balance and resource utilization. ing algorithm based on SLA and processing time suitable for cloud
environment,’’ Comput. Commun., vol. 151, pp. 183–195, Feb. 2020, doi:
At the same time, the TOPSIS-EWM algorithm is designed 10.1016/j.comcom.2019.12.050.
to select the best virtual machine for each task for multi- [10] N. Kumar, ‘‘A review on machine learning algorithms, tasks and appli-
objective workflow scheduling. It doesn’t taking into consid- cations,’’ Int. J. Adv. Res. Comput. Eng. Technol., vol. 6, no. 10, pp. 1–5,
Oct. 2017.
eration any user preferences like deadlines, load balance, and [11] K. Naik, G. Gandhi, and S. Patil, ‘‘Multi-objective virtual machine selec-
resource utilization. In contrast, the QL-HEFT algorithm for tion for task scheduling in cloud computing: ICCI-2017,’’ in Advances in
solving large-scale task optimization issues has limitations, Intelligent Systems and Computing. 2019, pp. 319–331.
[12] S. Pang, W. Li, H. He, Z. Shan, and X. Wang, ‘‘An EDA-GA
such as the Makespan task schedule. hybrid algorithm for multi-objective task scheduling in cloud
computing,’’ IEEE Access, vol. 7, pp. 146379–146389, 2019, doi:
10.1109/ACCESS.2019.2946216.
VIII. CONCLUSION AND FUTURE WORK
[13] M. S. Kumar, A. Tomar, and P. K. Jana, ‘‘Multi-objective workflow
In this paper, a new task scheduling algorithm using multi- scheduling scheme: A multi-criteria decision making approach,’’ J. Ambi-
objective based on a decision tree, called TS-DT algorithm, ent Intell. Hum. Comput., vol. 12, no. 12, pp. 10789–10808, Dec. 2021,
doi: 10.1007/s12652-020-02833-y.
is proposed for a cloud computing environment. The pro-
[14] J.-Q. Li and Y.-Q. Han, ‘‘A hybrid multi-objective artificial bee colony
posed TS-DT algorithm targets minimizing the makespan, algorithm for flexible task scheduling problems in cloud computing sys-
enhancing load balance, and maximizing resource utiliza- tem,’’ Cluster Comput., vol. 23, no. 4, pp. 2483–2499, Dec. 2020, doi:
tion. A comparative study was conducted to evaluate the 10.1007/s10586-019-03022-z.
[15] A. Kaur, P. Singh, R. S. Batth, and C. P. Lim, ‘‘Deep-Q learning-based
performance of the proposed TS-DT algorithm relative to the heterogeneous earliest finish time scheduling algorithm for scientific work-
HEFT, TOPSIS-EWM, and QL-HEFT algorithms. Accord- flows in cloud,’’ Softw. Pract. Exp., vol. 52, pp. 1–21, Jan. 2020, doi:
ing to the comparative results, the proposed TS-DT algorithm 10.1002/spe.2802.
[16] A. Al-maamari and F. A. Omara, ‘‘Task scheduling using PSO algorithm
outperforms the HEFT, TOPSIS-EWM, and QL-HEFT algo- in cloud computing environments,’’ Int. J. Grid Distrib. Comput., vol. 8,
rithms by reducing the average makespan by 5.21%, 2.54%, no. 5, pp. 245–256, Oct. 2015, doi: 10.14257/ijgdc.2015.8.5.24.
and 3.32%, respectively, improving the average resource [17] A. E. Keshk, ‘‘Cloud computing online scheduling,’’ IOSR J. Eng., vol. 4,
no. 3, pp. 07–17, Mar. 2014, doi: 10.9790/3021-04360717.
utilization by 4.69%, 6.81%, and 8.27%, respectively, and [18] J.-M. Yu, H.-H. Doh, Y.-J. Kwon, J.-H. Shin, H.-W. Kim, S.-H. Nam, and
improving the average load balancing by 33.36%, 19.69%, D.-H. Lee, ‘‘Decision tree based scheduling for static and dynamic flexible
and 59.06%. The main limitation of our proposed TS-DT job shops with multiple process plans,’’ J. Korean Soc. Precis. Eng., vol. 32,
no. 1, pp. 25–37, Jan. 2015, doi: 10.7736/KSPE.2015.32.1.25.
algorithm is the increase in power consumption has been [19] L. Yuan, Y. Dong, Y. Li, R. Zhang, and H. Xie, ‘‘A task parallel program-
increased by around 12.89%, 43.52%, and 28.38%, respec- ming framework based on heterogeneous computing platforms,’’ in Intelli-
tively, relative to the existing HEFT TOPSIS-EWM, and QL- gent Systems, Technologies and Applications. Jan. 2020, pp. 169–184, doi:
10.1007/978-981-15-3914-5_13.
HEFT algorithms. [20] S. K. Mishra, D. Puthal, B. Sahoo, S. K. Jena, and M. S. Obaidat, ‘‘An adap-
More performance parameters could be concerned in tive task allocation technique for green cloud computing,’’ J. Supercomput.,
future work, such as power consumption, fault tolerance, and vol. 74, no. 1, pp. 370–385, 2018, doi: 10.1007/s11227-017-2133-4.
[21] A. Dhari and K. I. Arif, ‘‘An efficient load balancing scheme for cloud
scalability. computing,’’ Indian J. Sci. Technol., vol. 10, no. 11, pp. 1–8, Mar. 2017,
doi: 10.17485/ijst/2017/v10i11/110107.
[22] M. S. Arif, Z. Iqbal, R. Tariq, F. Aadil, and M. Awais, ‘‘Parental
REFERENCES prioritization-based task scheduling in heterogeneous systems,’’
Arabian J. Sci. Eng., vol. 44, no. 4, pp. 3943–3952, Apr. 2019,
[1] J. Srinivas et al., ‘‘Cloud computing basics,’’ Creating Smart Enterprises,
doi: 10.1007/s13369-018-03698-2.
vol. 1, pp. 141–171, Jun. 2017, doi: 10.1201/9781315152455-6.
[23] K. Dubey, M. Kumar, and S. C. Sharma, ‘‘Modified HEFT algorithm
[2] D. Pooja, ‘‘Cloud computing—Overview and its challenges,’’ Int. J. Mul- for task scheduling in cloud environment,’’ Proc. Comput. Sci., vol. 125,
tidisciplinary, vol. 3085, no. 3, pp. 499–501, 2019. pp. 725–732, Jan. 2018, doi: 10.1016/j.procs.2017.12.093.
[3] M. Sajid and Z. Raza, ‘‘Cloud computing: Issues & challenges,’’ in Proc. [24] B. Singh and P. Mehta, ‘‘A survey of scheduling algorithms for heteroge-
Int. Conf. Cloud, Big Data Trust, Jun. 2015. neous systems and comparative study of HEFT and CPOP algorithms,’’
[4] Q. Jiang, Y. C. Lee, M. Arenaz, L. M. Leslie, and A. Y. Zomaya, ‘‘Opti- Int. J. Eng. Res., vol. V5, no. 5, pp. 250–254, May 2016.
mizing scientific workflows in the cloud: A montage example,’’ in Proc. [25] A. Mazrekaj, A. Sheholli, D. Minarolli, and B. Freisleben, ‘‘The expe-
IEEE/ACM 7th Int. Conf. Utility Cloud Comput., Dec. 2014, pp. 517–522, riential heterogeneous earliest finish time algorithm for task scheduling
doi: 10.1109/UCC.2014.77. in clouds,’’ in Proc. 9th Int. Conf. Cloud Comput. Services Sci., 2019,
[5] E. Sarhan, A. Ghalwash, and M. Khafagy, ‘‘Queue weighting load- pp. 371–379, doi: 10.5220/0007722203710379.
balancing technique for database replication in dynamic content web [26] Z. Xie, X. Shao, and Y. Xin, ‘‘A scheduling algorithm for cloud computing
sites,’’ Proc. 9th WSEAS Int. Conf. Appl. Comput. Sci. (ACS), Jan. 2009, system based on the driver of dynamic essential path,’’ PLoS One, vol. 11,
pp. 50–55. no. 8, pp. 1–19, 2016, doi: 10.1371/journal.pone.0159932.
[6] K. Kanagaraj and S. Swamynathan, ‘‘A realistic approach for repre- [27] B. A. Al-Maytami, P. Fan, A. Hussain, T. Baker, and P. Liatsis, ‘‘A task
senting and scheduling workflows in cloud computing environment,’’ in scheduling algorithm with improved makespan based on prediction of tasks
Proc. Int. Conf. Adv. Comput., Commun. Informat. (ICACCI), Sep. 2016, computation time algorithm for cloud computing,’’ IEEE Access, vol. 7,
pp. 1615–1621, doi: 10.1109/ICACCI.2016.7732279. pp. 160916–160926, 2019, doi: 10.1109/ACCESS.2019.2948704.
[7] H. Mahmoud, M. Thabet, M. H. Khafagy, and F. A. Omara, ‘‘A com- [28] Y. Samadi, M. Zbakh, and C. Tadonki, ‘‘E-HEFT: Enhancement hetero-
parative study of heterogenous task-based scheduling techniques in a geneous earliest finish time algorithm for task scheduling based on load
cloud environment,’’ in Proc. Int. Conf. Innov. Trends Commun. Com- balancing in cloud computing,’’ in Proc. Int. Conf. High Perform. Comput.
put. Eng. (ITCE), Feb. 2020, pp. 1–6, doi: 10.1109/ITCE48509.2020. Simul. (HPCS), Jul. 2018, pp. 601–609, doi: 10.1109/HPCS.2018.00100.
9047806. [29] Z. Tong, X. Deng, H. Chen, J. Mei, and H. Liu, ‘‘QL-HEFT: A novel
[8] H. M. Alkhashai and F. A. Omara, ‘‘An enhanced task scheduling algorithm machine learning scheduling scheme base on cloud computing environ-
on cloud computing environment,’’ Int. J. Grid Distrib. Comput., vol. 9, ment,’’ Neural Comput. Appl., vol. 32, no. 10, pp. 5553–5570, May 2020,
no. 7, pp. 91–100, Jul. 2016, doi: 10.14257/ijgdc.2016.9.7.10. doi: 10.1007/s00521-019-04118-8.

36150 VOLUME 10, 2022

H. Mahmoud et al.: Multiobjective Task Scheduling in Cloud Environment Using Decision Tree Algorithm

[30] H. Alinejad-Rokny, ‘‘Divide and conquer classification,’’ Austral. J. Basic MOSTAFA THABET received the Ph.D. degree in
Appl. Sci., vol. 5, no. 12, 2014. information systems, in 2019. He started M.B.A.
[31] K.-C. Jeong and Y.-D. Kim, ‘‘A real-time scheduling mechanism for a studies at the Arab Academy for Science Tech-
flexible manufacturing system: Using simulation and dispatching rules,’’ nology & Maritime Transport (AASTMT). He is
Int. J. Prod. Res., vol. 36, no. 9, pp. 2609–2626, 1998. currently a Lecturer with the Faculty of Comput-
[32] D. M. Abdelkader and F. Omara, ‘‘Dynamic task scheduling algorithm with ers and Information, Fayoum University, Egypt.
load balancing for heterogeneous computing system,’’ Egyptian Informat. He also works as a Media Consultant at Fayoum
J., vol. 13, no. 2, pp. 135–145, Jul. 2012, doi: 10.1016/j.eij.2012.04.001.
University for four years. He is a member of the
[33] H.-S. Choi, J.-S. Kim, and D.-H. Lee, ‘‘Real-time scheduling for reentrant
Big Data Research Group, Fayoum University.
hybrid flow shops: A decision tree based mechanism and its application
to a TFT-LCD line,’’ Expert Syst. Appl., vol. 38, no. 4, pp. 3514–3521, He has been working as the Vice President of the
Apr. 2011. Fayoum Cancer Center (FOC), since July 2020. He worked as a project man-
[34] N. Sasikaladevi, ‘‘Minimum makespan task scheduling algorithm in cloud ager of many projects at Fayoum University. His research interests include
computing,’’ Int. J. Grid Distrib. Comput., vol. 9, no. 11, pp. 61–70, the IoT, web, learning, database systems, high performance computing,
Nov. 2016, doi: 10.14257/ijgdc.2016.9.11.05. cloud computing, and big data. He worked as the Editor-in-Chief of Elfagr
[35] M. Pawlish, A. Varde, and S. Robila, ‘‘Analyze utilization rates in data Egyptian Newspaper. He is working as a Certified International Professional
centers to optimize energy management,’’ in Proc. Int. Green Comput. Trainer in a set of Egyptian organization.
Conf. (IGCC), 2012, pp. 1–6.
[36] T. Rashid and M. T. Rashid, ‘‘Design and implementation of load balancing
system for a smart home,’’ Proc. 3rd Int. Sci. Conf., Souther Tech. Univ.,
Mar. 2018, pp. 1–6.
[37] N. Quang-Hung and N. Thoai, ‘‘Minimizing total busy time with applica-
tion to energy-efficient scheduling of virtual machines in IaaS clouds,’’ in
Proc. Int. Conf. Adv. Comput. Appl. (ACOMP), Nov. 2016, pp. 141–148,
doi: 10.1109/ACOMP.2016.029.
[38] J. J. Kanet and V. Sridharan, ‘‘Scheduling with inserted idle time: Problem
taxonomy and literature review,’’ Oper. Res., vol. 48, no. 1, pp. 99–110,
Feb. 2000, doi: 10.1287/opre.48.1.99.12447.
[39] A. M. Chirkin, A. S. Z. Belloum, S. V. Kovalchuk, M. X. Makkes,
M. A. Melnik, A. A. Visheratin, and D. A. Nasonov, ‘‘Execution time
estimation for workflow scheduling,’’ Future Gener. Comput. Syst., vol. 75,
pp. 376–387, Oct. 2017, doi: 10.1016/j.future.2017.01.011. MOHAMED H. KHAFAGY received the Ph.D.
[40] V. Legout, M. Jan, and L. Pautet, ‘‘Scheduling algorithms to reduce the degree in computer science, in 2009. He worked
static energy consumption of real-time systems,’’ Real-Time Syst., vol. 51, as a Postdoctoral Researcher at the DIMA Group,
no. 2, pp. 153–191, Mar. 2015, doi: 10.1007/s11241-014-9207-7. Technique University Berlin, in 2012. He shares
[41] H. Zhao, G. Qi, Q. Wang, J. Wang, P. Yang, and L. Qiao, ‘‘Energy- to establish the first Big Data Research Group
efficient task scheduling for heterogeneous cloud computing sys- in Egypt with Cairo University, in 2013. He is
tems,’’ in Proc. IEEE 21st Int. Conf. High Perform. Comput. Com- currently the Manager of the Electronic Exams
munications; IEEE 17th Int. Conf. Smart City; IEEE 5th Int. Conf. Center, Supreme Council of Universities. He also
Data Sci. Syst. (HPCC/SmartCity/DSS), Aug. 2019, pp. 952–959, doi: works as a Consultant at Oracle Egypt. He is the
10.1109/HPCC/SmartCity/DSS.2019.00137. Head of the Big Data Research Group, Fayoum
[42] N. R. Ortiz-Pimiento and F. J. Díaz-Serna, ‘‘Relative average deviation University. He worked as a project manager of many projects at Fayoum Uni-
as measure of robustness in the stochastic project scheduling problem,’’ versity. He has many publications in the area of big data, cloud computing,
Revista Facultad Ingeniería, vol. 28, no. 52, pp. 77–97, Jun. 2019, doi:
and database systems.
10.19053/01211129.v28.n52.2019.9756.
[43] R. N. Calheiros, R. Ranjan, A. Beloglazov, C. A. F. De Rose, and R. Buyya,
‘‘Cloudsim: A toolkit for modeling and simulation of cloud computing
environments and evaluation of resource provisioning algorithms,’’ Softw.,
Pract. Exper., vol. 41, no. 1, pp. 23–50, 2011.
[44] Caltech IPAC Software. Accessed: May 2021. [Online]. Available:
https://github.com/Caltech-IPAC/MontageMosaics
[45] E. Zitzler, L. Thiele, M. Laumanns, C. M. Fonseca, and V. G. da Fonseca,
‘‘Performance assessment of multiobjective optimizers: An analysis and
review,’’ IEEE Trans. Evol. Comput., vol. 7, no. 2, pp. 117–132, Apr. 2003.

HADEER MAHMOUD received the bachelor’s

degree in computer science from the Faculty FATMA A. OMARA (Member, IEEE) is cur-
of Information Systems and Computer Science, rently a Professor with the Department of Com-
October 6 University, Egypt, in 2013, and the puter Science, Faculty of Computers and Artificial
M.Sc. degree in computer science from the Col- Intelligence, Cairo University. She has supervised
lege of Computing and Information Technology, over 50 Ph.D. and M.Sc. theses. She has published
Arab Academy for Science Technology and Mar- over 100 research papers in prestigious interna-
itime Transport, in 2018. She is currently pursuing tional journals and conference proceedings. Her
the Ph.D. degree with the Faculty of Computers research interests include parallel and distribut-
and Information, Fayoum University, Egypt. She is ing computing, distributed operating systems, high
currently a Staff Member with the Department of Computer Science, Faculty performance computing, grid, cloud computing,
of Information Systems and Computer Science, October 6 University. She and big data. She is a member of the IEEE Computer Society. She has
is also a Cloud Computing Instructor at Huawei Academy. Her research served as the chairman and a member of Steering Committees and Program
interests include high-performance computing, cloud computing, big data, Committees of several national conferences.
and security.

VOLUME 10, 2022 36151