Content uploaded by Tawfiq Alrawashdeh
Author content
All content in this area was uploaded by Tawfiq Alrawashdeh on Feb 13, 2019
Content may be subject to copyright.
Copyright © 2018 Authors. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.
International Journal of Engineering & Technology, 7 (3.28) (2018) 271-274
International Journal of Engineering & Technology
Website: www.sciencepubco.com/index.php/IJET
Research paper
Scientific Workflow Scheduling in Clouds: A Review
Tawfiq Alrawashdeh1, Aznida Hayati Zakaria2*, Zarina Mohamad2
1Al Husein Bin Talal University, P.O. Box 20 Ma'an, Jordan
2Faculty of Informatics and Computing, Universiti Sultan Zainal Abidin, Besut Campus, Terengganu, Malaysia
*Corresponding author E-mail: aznida@unisza.edu.my
Abstract
Due to their abundant resources that can be elastically provisioned with pay-as-you-go pricing, clouds have emerged as a promising cost-
efficient platform to execute large scale scientific applications. Such applications consist of number of processes/tasks forming workflow.
These tasks are connected by direct edges that show the data dependency between the tasks. Tasks perform their computation on the
original data submitted by the user, or on data passed by its predecessor task. This work, classify and discuss proposals that investigate
the problem of scheduling scientific workflows in clouds.
Keywords: scheduling algorithms; scientific workflow; cloud computing.
1. Introduction
With the rapid increment on the complexity of the workflow, and
the resultant demand on the scalability of the environment, execut-
ing workflows on traditional environment become a very challeng-
ing task. To address this issue, cloud computing (especially Infra-
structure as a Service) has emerge as an efficient environment to
execute scientific workflows.
Infrastructure as a Service (IaaS) cloud offer an effectively availa-
ble, adaptable, and versatile foundation for the arrangement of
scientific workflows. Using IaaS, user can rent Virtual Machines
(VMs) to execute their computational tasks. This allows user ac-
cess to a practically unbounded pool of VMs that can be flexibly
gained and discharged during the execution of the computational
task(s). In this service, users are charged based on the number of
resources they rent using pay-per-use cost model.
In this direction, to increase the utilization of the resources, must
determine the right number of resources to rent. Over-renting is
expected to increase the execution cost, and under-renting is ex-
pected to reduce the performance (increase execution time). This
problem is underlined due to the bi-objective scheduling problem
presented by scheduling workflows in cloud. Generally, want to
establish an execution schedule for the tasks of the workflows
such that the execution cost and time is minimized. This conflict
in the objective results in highlighting the process of determine the
number of resources to rent as a major challenge.
The problem of scheduling workflows in clouds is NP-Complete
in nature. Many variations of this problem have been proposed in
the literature. For instance, many proposals have investigated the
problem of finding the cheapest schedules that satisfy pre-
determined deadline on the execution of the workflow [1-5, 7]. On
the other hand, in [6, 9-11], the authors addressed the problem of
determining the schedule of executing the tasks such that the exe-
cution time is minimized and pre-determined execution cost con-
straint is satisfied. Introducing budget or/and deadline constraints
results in reducing the optimization space to one of the objectives
(cost, time).
This paper, discuss and classify proposals that investigate the
problem of scheduling scientific workflows in clouds. Section 2
presents the most related approaches to this problem, and in sec-
tion 3, conclude this paper.
2. Scheduling Approaches
Many proposals [1-24] have investigated the problem of
scheduling scientific workflows in cloud. In this section,
this paper discusses the related proposals in the literature.
This paper categorizes these proposals in term of their ob-
jective function into: (1) single-objective, and (2) multi-
objective.
2.1. Single-objective
The problem of scheduling scientific workflows on with the objec-
tive of minimizing the makespan or cost minimization has been
studied extensively in the literature. For instance, given a pre-
determined budget-constraint, in [1] propose a heuristic-based
solution that aim to reduce the overall execution delay. In each
iteration, the main idea of this approach is to improve the current
schedule by considering the left budget. In [3] addressed the prob-
lem of minimizing the makespan under the presence of budget
constraints. They termed this problem as the Minimum End-to-end
Delay under Cost Constraint (MED-CC) problem. The authors
addressed the hardness of this problem by proving that it is NP-
Complete, and it is non-approximable. To address this problem,
the authors proposed heuristic-based solution. This heuristic fol-
lows an efficient searching strategy, where in each iteration it tries
to find a new schedule such that the makespan is minimized.
In [4] proposed priority based genetic algorithm termed as
BCHGA, which address the problem of scheduling the tasks of the
workflow under the presence of the budget constraint. In this algo-
rithm, based on the locality of the tasks, each task is assigned ei-
ther bottom level priority (b-level), or top-level priority (t-level).
Then, in each round, the algorithm by trying to find better sched-
272
International Journal of Engineering & Technology
ule in term of makespan, while minimizing the execution cost. In
[2, 5], the authors adopt similar strategy to minimize the execution
cost under the presence of the budget and execution time con-
straints.
In [6] proposed a Deadline guarantee enhanced scheduling algo-
rithm (DGESA). This algorithm target scenario where scientific
workflows can be scheduled on hybrid. This algorithm starts by
calculating the sub-deadline for each task. For each task, once the
sub-deadline is calculated, the algorithm proceeds by determining
the probability of violating thi s deadline. For each task, if the
probability of violating the deadline is higher than pre-determined
threshold, this task will be scheduled to be executed on public
cloud. Otherwise, the task will be executed on private cloud.
In [7] also addressed the problem of minimizing the execution
time under the presence of the budget constraint. To address this
problem, the authors' proposed budget-driven approach, which
start by partitioning the workflow into bags of tasks, where each
bag contains a group of parallel tasks. Then each bag is scheduled
based on the characteristics and locality of its tasks. This is estab-
lished by modelling the resource provisioning plan for each bag of
tasks as a mixed integer linear programming (MILP) model.
In [8] proposed a new pulling-based workflow execution system
with a profiling-based resource provisioning strategy (DEWE v2).
This system consists of master and worker daemons, which can be
run by the same node or different nodes. The master daemon is
responsible for managing the progress of executing the workflows.
The worker daemon informs the master daemon whenever a task
has been successfully executed. In addition, this system has work-
er submission application, which can be used by the scientist to
submit workflows
In [9], the authors proposed the IC-PCP algorithm that to schedule
the workflow's tasks such the execution cost is minimized and a
pre-determine execution deadline is satisfied. This algorithm starts
by distributing the workflow deadline across the entire workflows
tasks. Starting from the critical path tasks, the sub-deadline for
these path tasks will be determined. Then, the algorithm proceeds
by determining the deadlines for all sub-paths. Once the deadline
for each task is determined, each task will be schedule on the re-
source that result in satisfying the task deadline and reducing the
execution cost.
In [10], the authors propose scheduling strategy designed mainly
for WaaS, which aim to minimize the execution cost while satisfy-
ing a pre-determined time-deadline. The main idea of this algo-
rithm is to determine each task sub-deadline. Then, once a task
become ready for execution, if no available VM can be used to
schedule this task without violating its deadline, a new VM will be
leased.
In [11], the authors proposed an approximation algorithm to ad-
dress the problem of scheduling scientific workflows such that the
execution cost is minimized, while satisfying a pre-determined
time deadline. This algorithm works in iterative fashion, where in
each iteration the scheduler tries to find the maximum number of
tasks that can be executed at a specific node. This is established by
considering the latest acceptable finishing time for each task.
2.2. Multi-objective
Now, this paper discusses in details proposals that deal with bi-
objective problems. This paper focus on the problem of minimiz-
ing the execution cost and time. This problem has conflicted ob-
jective, since reducing the makespan results in increasing the cost,
and reducing the cost result in increasing the makespan.
In [12], proposed solution that address the problem of minimizing
both the execution cost and execution time. Their approach can be
considered as an extended version of the well-known Heterogene-
ous Earliest Finish Time (HEFT) heuristic, and it attempt to estab-
lish a trade-off between execution time and execution cost. This
main idea of this heuristic is to assign a priority value for each
task. This priority value is determine based on the locality of each
task, and it aim to prioritize executing tasks on VMs such that
executing cost and time is minimized.
In [13], the Deadline-Budget Workflow Scheduling (DBWS) algo-
rithm is presented to address the problem of scheduling scientific
workflows in clouds, while the execution cost and time are mini-
mized. This algorithm assume that the user specifies pre-
determined time and cost deadlines. Makespan is handled as a
primary objective and thus the resultant schedule must always
satisfy the deadline constraint. In this algorithm, cost is handled as
a secondary objective, where the cost constraint can be violated.
At its core, this algorithm tries to find the cheapest schedule that
can satisfy the deadline constraint.
In [14], the authors proposed dynamic auction-based workflow
scheduling algorithm that dynamically allocate the workflow's
tasks across multiple cloud domains. The objective of this algo-
rithm is to reduce the execution cost, and satisfy the execution
time constraint. In this algorithm, each task is ranked based on its
level load, and its successor rank. In this algorithm, task with
submit their input and output requirement, and resources will bid
their computational capacity. In this paradigm, tasks will be as-
signed to resources with higher computational capacity.
In [15] consider a multi-cloud domain, where every supplier con-
tributes a fixed number of heterogeneous VMs. In addition, they
provide global storage service to store intermediate data files. The
authors formulate the scheduling problem as a Mixed Integer Pro-
gram (MIP). Then they proposed two algorithmic solutions to
address two different scenarios. In the first scenario, they assume
that the running time for each is in hour unit, where in the second
scenario they assume that each task running time is less than an
hour. They start by partitioning the tasks based on their level. In
addition, they assumed that any VMs cannot be allocated different
partitions. This simplify the execution of the proposed MIP. How-
ever, by limiting the solution space, this approach reduces the
optimization space. In this direction, in [16] presented a clustering
approach to schedule workflows in clouds, where the tasks levels
are the key factor behind the schedule.
In [17] presented a dynamic resource provisioning and scheduling
algorithm DPDS. This algorithm aims to schedule multiple work-
flows simultaneously on the cloud. Such schedule is established
with the objective of satisfying the users cost and time constraints.
It attempts to maximize the number of completed workflows, and
does not take into account minimizing the execution cost. Fur-
thermore, it assumes that the available resources have the same
computational and communication capabilities.
In [18] presented the Workflow scheduling on Hybrid Cloud to
maintain Data Privacy (WHPD) algorithm. The objective of this
algorithm is to maintain the privacy of the scheduled tasks such
that the makespan constraint is satisfied. This algorithm handled
minimizing the execution cost as a secondary objective. This algo-
rithm starts by determining the latest starting time for each task, in
order to satisfy the pre-determined time-deadline. Then, each task
will be allocated to the VM such that the gap between the task
earliest execution time and the actual execution time is minimized.
The last step of this algorithm is to attempt improving the obtained
schedule by moving tasks between VMs in order to reduce the
execution time. The performance of this algorithm depends on the
initial schedule since, this schedule establishes constraints on the
allowed movements in the schedule improvement step.
In [19], the authors proposed the Cost with Finish Time-Based
(CwFT) Algorithm. This algorithm is an extended version of the
HEFT algorithm, and it aim to reduce the execution time and cost.
This algorithm consists of two phases: (1) task prioritizing and (2)
node selection. In the task prioritizing phase, the priority of each
task will be determined based on its locality. In the node selection
phase, each task will be assigned to the VM with the objective of
minimizing that ratio of executing cost and time. In [20], the au-
thors investigated the same problem, and they also proposed an
auto-scaling algorithm, which adopt the strategy of the well-
known HEFT algorithm.
Several authors [21-24] have investigated the problem of schedul-
ing scientific workflows in cloud using nature-inspired algorithm.
Although techniques like Ant Colony Optimization (ACO) and
Particle Swarm Optimization (PSO), and Genetic Algorithms
International Journal of Engineering & Technology
273
(GA), can lead to find a near optimal solution, its main issue is
practicality. Such techniques may require relatively large pro- cessing time, and this reduces the scalability of these approaches.
Table 1: Summary table of related work on workflow scheduling
References
Algorithm
Environment Tools
Optimization Strategy
Constraints
[1]
Critical-Greedy(CG)
CloudSim
Heuristic
single-objective
[2]
BPSO
Synthetic workflows
Heuristic
single-objective
[3]
MED-CC
CloudSim
Heuristic
single-objective
[4]
BCHGA
Java
Metaheuristic
single-objective
[5]
PDC and DCCP
Compared to existing algorithms
Heuristic
single-objective
[6]
DGESA
Simulation environment of hybrid computing
Heuristic
single-objective
[7]
BAGS
Synthetic workflows
Heuristic
single-objective
[8]
DEWE v2
Pegasus
Metaheuristic
single-objective
[9]
IC-PCP and IC-PCPD2
Synthetic workflows
Heuristic
single-objective
[10]
EPSM
Synthetic workflows
Heuristic
single-objective
[11]
EES
Synthetic workflows
Heuristic
single-objective
[12]
BDHEFT
Synthetic workflows
Heuristic
multi-objective
[13]
DBWS
Synthetic workflows
Heuristic
multi-objective
[14]
Novel replication aware
dynamic workflow scheduling
CloudSim toolkit
Heuristic
multi-objective
[15]
Mathematical models
Amazon EC2
Heuristic
multi-objective
[16]
Efficient workflow
scheduling algorithm
WorkflowSim and CloudSim
Heuristic
multi-objective
[17]
DPDS
CloudSim
Heuristic
multi-objective
[18]
WHPD
Synthetic workflows
Heuristic
multi-objective
[19]
CwFT
Gaussian Elimination program
Heuristic
multi-objective
[20]
HEFT
Compare with approaches
Heuristic
multi-objective
[21]
Algorithm based on the
meta-heuristic optimization technique
Synthetic workflows
Metaheuristic
multi-objective
[22]
CEGA
Synthetic workflows
Metaheuristic
multi-objective
[23]
A combined resource provisioning
and scheduling strategy
CloudSim framework
Metaheuristic
multi-objective
[24]
A hybrid genetic algorithm
WorkflowSim
Metaheuristic
multi-objective
3. Conclusion
This paper discussed the problem of scheduling scientific work-
flow in clouds under the presence of budget and/or deadline con-
straints. Based on the objective function of the problem, this paper
is mainly interested in three variations of this problem. In the first
problem, discussed proposals related to the problem of minimizing
the execution cost. In this problem, the users are typically main
concern with the cost of the execution and have no restriction on
the execution time. In the second problem, focus on minimizing
the execution time. In this problem, users target minimizing the
makespan, and this may lead to increase the execution cost. In the
last problem, focus on proposals related to minimizing the execu-
tion cost and time. In this problem users try to find a solution that
balance the importance of time and cost.
References
[1] Wu, C., Lin, X., Yu, D., Xu, W., & Li, L. (2015). End-to-end delay
minimization for scientific workflows in clouds under budget
constraint. IEEE Transactions on Cloud Computing, 3(2), 169-181.
[2] Verma, A., & Kaushal, S. (2015). Cost minimized PSO based
workflow scheduling plan for cloud computing. I.J. Information
Technology and Computer Science, 8, 37-43.
[3] Lin, X., & Wu, C. Q. (2013, October). On scientific workflow
scheduling in clouds under budget constraint. Proceedings of the
IEEE 42nd International Conference on Parallel Processing, pp. 90-
99.
[4] Verma, A., & Kaushal, S. (2013). Budget constrained priority
based genetic algorithm for workflow scheduling in cloud.
Proceedings of the IET Fifth International Conference on Recent
Trends in Information, Telecommunication and Computing, pp. 8-
14.
[5] Arabnejad, V., Bubendorfer, K., & Ng, B. (2017). Scheduling
deadline constrained scientific workflows on dynamically
provisioned cloud resources. Future Generation Computer
Systems, 75, 348-364.
[6] Luo, H., Yan, C., & Hu, Z. (2015). An Enhanced Workflow
Scheduling Strategy for Deadline Guarantee on Hybrid Grid/Cloud
Infrastructure. Journal of Applied Science and Engineering, 18(1),
67-78.
[7] Rodriguez, M. A., & Buyya, R. (2017). Budget-driven scheduling
of scientific workflows in IaaS clouds with fine-grained billing
periods. ACM Transactions on Autonomous and Adaptive
Systems, 12(2), 1-22.
[8] Jiang, Q., Lee, Y. C., & Zomaya, A. Y. (2015). Executing large
scale scientific workflow ensembles in public clouds. Proceedings
of the IEEE 44th International Conference on Parallel Processing,
pp. 520-529.
[9] Abrishami, S., Naghibzadeh, M., & Epema, D. H. (2013).
Deadline-constrained workflow scheduling algorithms for
infrastructure as a service clouds. Future Generation Computer
Systems, 29(1), 158-169.
[10] Rodriguez, M. A., & Buyya, R. (2018). Scheduling dynamic
workloads in multi-tenant scientific workflow as a service
platforms. Future Generation Computer Systems, 79, 739-750.
[11] Ma, Y., Gong, B., Sugihara, R., & Gupta, R. (2012). Energy-
efficient deadline scheduling for heterogeneous systems. Journal of
Parallel and Distributed Computing, 72(12), 1725-1740.
[12] Verma, A., & Kaushal, S. (2015). Cost-time efficient scheduling
plan for executing workflows in the cloud. Journal of Grid
Computing, 13(4), 495-506.
[13] Ghasemzadeh, M., Arabnejad, H., & Barbosa, J. G. (2017).
Deadline-budget constrained scheduling algorithm for scientific
workflows in a cloud environment. Proceedings of the LIPIcs-
Leibniz International Proceedings in Informatics, pp. 1-16.
[14] Gayathri, T., & Subashini, B. V. (2015). Task ranking based
allocation of scientific workflows in multiple clouds with deadline
constraint. International Journal of Engineering and Computer
Science, 4(2), 10543-10546.
[15] Malawski, M., Figiela, K., Bubak, M., Deelman, E., & Nabrzyski, J.
(2015). Scheduling multilevel deadline-constrained scientific
workflows on clouds based on cost optimization. Scientific
Programming, 2015, 1-13.
[16] Prathibha, D. A., Latha, B., Sumathi, G., Vani, R., Sangeetha, M.,
Davis, P., Nithyanandam C, Mohankumar G, Suratanee A, Lertsari
N, & Kamphasee, S. (2014). Efficient scheduling of workflow in
cloud environment using billing model aware task
clustering. Journal of Theoretical and Applied Information
Technology, 65(3), 595-605.
[17] Malawski, M., Juve, G., Deelman, E., & Nabrzyski, J. (2015).
Algorithms for cost-and deadline-constrained provisioning for
scientific workflow ensembles in IaaS clouds. Future Generation
Computer Systems, 48, 1-18.
274
International Journal of Engineering & Technology
[18] Abrishami, H., Rezaeian, A., & Naghibzadeh, M. (2015).
Workflow scheduling on the hybrid cloud to maintain data privacy
under deadline constraint. Journal of Intelligent Computing, 6(3),
92-103.
[19] Man, N. D., & Huh, E. N. (2013). Cost and efficiency-based
scheduling on a general framework combining between cloud
computing and local thick clients. Proceedings of the IEEE
International Conference on Computing, Management and
Telecommunications, pp. 258-263.
[20] Jiping, Z., Chunhua, G., & Feng, W. (2014). HEFT based cloud
auto-scaling algorithm with budget constraints. International
Journal of Advances in Computer Science and Technology, 3, 13-
18.
[21] Goyal, M., & Aggarwal, M. (2017). Optimize workflow scheduling
using hybrid ant colony optimization (ACO) and particle swarm
optimization (PSO) algorithm in cloud environment. International
Journal of Advance Research, Ideas and Innovations in Technology,
3(2), 1-9.
[22] Meena, J., Kumar, M., & Vardhan, M. (2016). Cost effective
genetic algorithm for workflow scheduling in cloud under deadline
constraint. IEEE Access, 4, 5065-5082.
[23] Rodriguez, M. A., & Buyya, R. (2014). Deadline based resource
provisioning and scheduling algorithm for scientific workflows on
clouds. IEEE Transactions on Cloud Computing, 2(2), 222-235.
[24] Kaur, G., & Kalra, M. (2017). Deadline constrained scheduling of
scientific workflows on cloud using hybrid genetic algorithm.
Proceedings of the IEEE 7th International Conference on Cloud
Computing, Data Science and Engineering-Confluence, pp. 276-
280.
