CC BY
25
НАУЧНО-ТЕХНИЧЕСКИЙ ВЕСТНИК ИНФОРМАЦИОННЫХ ТЕХНОЛОГИЙ, МЕХАНИКИ И ОПТИКИ март-апрель 2023 Том 23 № 2 http://ntv.ifmo.ru/
I/ITMO SCIENTIFIC AND TECHNICAL JOURNAL OF INFORMATION TECHNOLOGIES, MECHANICS AND OPTICS ИНФОРМАЦИОННЫХ ТЕХНОЛОГИЙ, МЕХАНИКИ И ОПТИКИ
March-April 2023 Vol. 23 No 2 http://ntv.ifmo.ru/en/
ISSN 2226-1494 (print) ISSN 2500-0373 (online)
10.17586/2226-1494-2023-23-2-313-322
Hybrid JAYA algorithm for workflow scheduling in cloud
Sandeep Kumar Bothra1«, Sunita Singhal2, Hemlata Goyal3
1,2,3 Manipal University Jaipur, Jaipur, Rajasthan, 303007, India
1 bothrajain@gmail.com«, https://orcid.org/0000-0003-0555-569X
2 sunita.singhal@jaipur.manipal.edu, https://orcid.org/0000-0003-2462-8102
3 hemlata.goyal@jaipur.manipal.edu, https://orcid.org/0000-0003-1344-0921
Abstract
Workflow scheduling and resource provisioning are two of the most critical issues in cloud computing. Developing an optimal workflow scheduling strategy in the heterogeneous cloud environment is extremely difficult due to its NP-complete nature. Various optimization algorithms have been used to schedule the workflow so that users can receive Quality of Service (QoS) from cloud service providers as well as service providers can achieve maximum gain but there is no such model that can simultaneously minimize execution time and cost while balancing the load among virtual machines in a heterogeneous environment using JAYA approach. In this article, we employed the hybrid JAYA algorithm to minimize the computation cost and completion time during workflow scheduling. We considered the heterogeneous cloud computing environment and made an effort to evenly distribute the load among the virtual machines. To achieve our goals, we used the Task Duplication Heterogeneous Earliest Finish Time (HEFT-TD) and Predict Earliest Finish Time (PEFT). The makespan is greatly shortened by HEFT-TD which is based on the Optimistic Cost Table. We used a greedy technique to distribute the workload among Virtual Machines (VMs) in a heterogeneous environment. Greedy approach assigns the upcoming task to a VM which have lowest load. In addition, we also considered performance variation, termination delay, and booting time of virtual machines to achieve our objectives in our proposed model. We used Montage, LIGO, Cybershake, and Epigenomics datasets to experimentally analyze the suggested model in order to validate the concept. Our meticulous experiments show that our hybrid approach outperforms other recent algorithms in minimizing the execution cost and makespan, such as the Cost Effective Genetic Algorithm (CEGA), Cost-effective Load-balanced Genetic Algorithm (CLGA), Cost effective Hybrid Genetic Algorithm (CHGA), and Artificial Bee Colony Algorithm (ABC). Keywords
JAYA algorithm, workflow scheduling, execution cost, makespan, load balance, cloud computing
For citation: Bothra S.K., Singhal S., Goyal H. Hybrid JAYA algorithm for workflow scheduling in cloud. Scientific
and Technical Journal of Information Technologies, Mechanics and Optics, 2023, vol. 23, no. 2, pp. 313-322. doi:
10.17586/2226-1494-2023-23-2-313-322
УДК 004.75
Гибридный алгоритм JAYA для планирования рабочих процессов в облаке
Сандип Кумар Ботра1Н, Сунита Сингхал2, Хемлата Гоял3
1,2,з Манипальский университет Джайпура, Джайпур, Раджастхан, 303007, Индия
1 bothrajain@gmail.coms, https://orcid.org/0000-0003-0555-569X
2 sunita.singhal@jaipur.manipal.edu, https://orcid.org/0000-0003-2462-8102
3 hemlata.goyal@jaipur.manipal.edu, https://orcid.org/0000-0003-1344-0921
Аннотация
Планирование рабочих процессов и предоставление ресурсов — две наиболее важные проблемы облачных вычислений. Разработка оптимальной стратегии планирования рабочих процессов в гетерогенной облачной среде чрезвычайно сложна из-за ее NP-полной природы. При планировании рабочего процесса используются различные алгоритмы оптимизации для получения пользователями качественного обслуживания (Quality of Service, QoS) от поставщиков облачных услуг. При этом поставщики услуг должны получать максимальную выгоду. Сегодня не существует такой модели, которая могла бы одновременно минимизировать время и
© Bothra S.K., Singhal S., Goyal H., 2023
стоимость выполнения работ при балансировке нагрузки между виртуальными машинами в гетерогенной среде с использованием подхода JAYA. В работе предложен гибридный алгоритм JAYA для минимизирования стоимости вычислений и времени выполнения работ при планировании рабочего процесса. Рассмотрена гетерогенная среда облачных вычислений, позволяющая равномерно распределять нагрузку между виртуальными машинами. Для достижения этих целей использованы эвристические подходы Task Duplication Heterogeneous Earliest Finish Time (HEFT-TD) и Predict Earliest Finish Time (PEFT). Длительность выполнения работ значительно сокращается благодаря HEFT-TD, основанному на таблице оптимистичных затрат (Optimistic Cost Table). Для распределения рабочей нагрузки между виртуальными машинами в гетерогенной среде использован жадный алгоритм. Жадный алгоритм назначает предстоящую задачу виртуальной машине с наименьшей нагрузкой. Рассмотрено изменение производительности, задержки завершения и время загрузки виртуальных машин. С целью проверки предложенной концепции для экспериментального анализа представленной модели использованы наборы данных Montage, LIGO, Cybershake и Epigenomics. Выполненные эксперименты показали, что рассмотренный гибридный подход превосходит более ранние алгоритмы по минимизации стоимости и времени его выполнения, такие как Cost Effective Genetic Algorithm (CEGA), Cost-effective Load-balanced Genetic Algorithm (CLGA), Cost effective Hybrid Genetic Algorithm (CHGA) и Artificial Bee Colony Algorithm (ABC). Ключевые слова
алгоритм JAYA, планирование рабочего процесса, стоимость выполнения, время выполнения, баланс нагрузки, облачные вычисления
Ссылка для цитирования: Ботра С.К., Сингхал С., Гоял Х. Гибридный алгоритм JAYA для планирования рабочих процессов в облаке // Научно-технический вестник информационных технологий, механики и оптики. 2023. Т. 23, № 2. С. 313-322 (на англ. яз.). doi: 10.17586/2226-1494-2023-23-2-313-322
Introduction
Cloud computing is an emerging sector of computing in which a collection of resources is provided to a user as a service rather than as a product. The best thing about these services is that consumers do not need to be aware of the actual locations of the resources or the configurations of these resources that provide the needed service. A workflow submitted by a user to be executed on the cloud is a group of interdependent tasks that are represented by a Directed Acyclic Graph [1]. When workflow scheduling is done in a heterogeneous computing system, the problem becomes more complicated since the processors in the distributed environment may not be similar and require varying amounts of time to complete the same operation. Workflow scheduling and resource provisioning are two of the most critical issues in cloud computing. Developing an optimal workflow scheduling technique in the heterogeneous cloud environment is extremely difficult due to its NP-complete nature [2].
Various optimization algorithms [3] have been used to schedule the workflow so that users can receive Quality of Service (QoS) from cloud service providers. Task scheduling is essential for optimum utilization of cloud resources and also for providing end-users with a QoS [4]. Task scheduling issues come in two flavors: static scheduling and dynamic scheduling. In the static category, all task details, including the costs of computation and communication for each activity as well as how those activities relate to one another, are known in advance. However, in the dynamic category, such data is not available, and choices are made in real time. Furthermore, static scheduling refers to compile-time scheduling, and dynamic scheduling refers to scheduling at runtime [5].
Heuristic approaches are problem-dependent and frequently too greedy, resulting in their becoming stuck in a local optimum and failing to achieve an optimal solution. When there is partial information or limited computing power, a meta-heuristic is a higher-level technique or heuristic that is used to find, generate, or select a heuristic
capable of offering a good solution to an optimization problem.
Due to the poor convergence rate of the meta-heuristic technique, achieving an optimal solution is difficult. As a result, a one-size-fits-all solution will not guarantee optimal resource utilization. That's why, the hybrid nature of the approach is one of the best ways to reach our goal function. So we developed a new model using the JAYA algorithm where seeding is performed using the Task Duplication Heterogeneous Earliest Finish Time (HEFT-TD) [6] and Predict Earliest Finish Time (PEFT) [7] heuristics. We made an effort in this paper to shorten the computation time and cost while still meeting the deadline.
Our goal was to create a hybrid metaheuristic strategy that was effective for decreasing processing time and costs while maintaining load balance amongst Virtual Machines (VMs) under time restrictions. The strategy used in this work employs a HEFT-TD and PEFT strategies during population initialization, which helps with cost cutting and load balancing.
The following is the rest of the article which summarizes prior research in this topic. The implementation and outcomes of the recommended method, as well as a comparison with existing methods, are discussed in further detail. This publication also includes a discussion and conclusion that recommends additional research in this area.
Related Work
The objective of our model in a cloud computing context is to improve the system turnaround time and resource usage. The number of interdependent jobs and available resources in a distributed environment varies dynamically. Various scheduling algorithms developed to obtain optimum solution, some of them heuristic and other are meta-heuristic as both methodologies are integral to maintaining optimal resource scheduling as we mentioned earlier.
Based on the Particle Swarm Optimization (PSO) technique, the authors develop a multi-objective algorithm
for scheduling workflow. They use two factors in their innovative method to fulfill their goals: makespan and resource usage as well as a strict encoding technique. Despite the fact that their testing results show that their technique is more resilient than baseline approaches, they fail to account for VMs balancing [8].
The authors [9] use a Pareto distribution to allocate unused VMs in Ant Colony Optimization (ACO) to minimize computation cost and time. They also use a minimal relocation of VMs strategy to boost the effectiveness of their approach in assessing workflow computation time and price; however, as their approach is dependent on a relatively small workflow size, the performance of the algorithm is uncertain.
To schedule the workflow, the authors enhanced PSO. In order to achieve the lowest operational time and cost, they started by regulating the global and local performance of particles using a nonlinear decreasing technique of inertia weight. However, they failed to take into account the dynamic nature of the cloud computing environment [10].
An Artificial Bee Colony (ABC) based algorithm is proposed in the literature [11], with the authors emphasizing the QoS regulations and critical security principles. A hive table is kept in a data center to reduce execution cost, execution time, job migration, and VM load-balancing. They should build a hybrid strategy because the ABC approach alone is unable to manage all of these aspects.
For load balancing during workflow scheduling in the cloud, the author [12] uses the JAYA algorithm. Authors [13] suggested a modified backfilling technique for optimal cloud resource use in which they schedule jobs without a decision-maker to firmness conflicts. To lower the execution cost and duration, a vocalization of the humpback whale optimization algorithm [14] is presented. By using less energy, this technique protects the environment. Authors [15] proposed an efficient method for scheduling work in the cloud utilizing the MAPREDUCE and GA-WOA. The proposed approach consists of stages, such as feature reduction, feature selection, task separation, and task scheduling. They intended to reduce makespan, however they did not consider load balance among virtual machines.
Although the authors [16] integrated execution time and throughput in their model based on Bat algorithm, but they didn't take into account communication time which is a crucial component in reducing execution time and increasing throughput. They were likewise unconcerned about the load balance between the multiple VMs also. This obscurity is eliminated in [17] when the authors tried to optimize the resource allocation for VMs and suggested using the Bat approach to evenly distribute the workload over several VMs. In the paper [18], authors proposed a hybrid method that combined Heterogeneous Earliest Finish Time (HEFT) and Genetic Algorithm (GA) to reduce processing costs and time under budget limitations; however, they were unable to account for VM booting time. In the literature [19], a multi-objective load balancing method, which is based on GA, was proposed; in it delays in acquisition are ignored while cost and time are decreased. In the literature [20], authors described an approach utilizing JAYA in which they only concentrated
on minimizing the execution cost and makespan but did not take into account the performance variance and acquisition delay of the VM as well as they did not aware regarding load balance among VMs. To address these issues, we were inspired to implement a hybrid JAYA model "HJA".
After a thorough examination of the literature, we determined the research gap that researchers' applied heuristic methods to workflow scheduling are ineffective due to the NP-hard nature of the problem. Iteration is required for metaheuristic techniques to reach an optimal solution. As a consequence, hybrid metaheuristic approaches outperform traditional metaheuristic techniques in terms of identifying the optimal solution.
Proposed Methodology
Description of JAYA
After a deep literature review, we decided to implement a hybrid model using the JAYA algorithm (HJA) which is based on the heuristics of HEFT-TD and PEFT. Since all evolutionary and swarm cognitive algorithms are unpredictable and necessitate the same governing variables, such as size of population and iteration number, we choose the JAYA [21] technique. In addition to the standard control parameters, various algorithms, such as ACO, PSO, GA, etc., also need their own algorithm-specific parameter settings, such as pheromone value, evaporation rate, cognitive acceleration constant, and crossover rate. The above-mentioned algorithms performance is greatly impacted by the right adjustment of algorithm-specific parameters. Inadequately adjusting algorithm-specific parameters either leads to the best local result or increases processing effort. There are no algorithm-specific parameters needed for JAYA. It was invented by R. Venkata Rao in 2016 [21]. It is a parameter-less algorithm that requires a few iterations to achieve an optimum solution.
Xjtk1 = Xk + Thibet - Xjkl) - r2jXj!worst - Xjki). (1)
New solution is obtained by applying the above mentioned eq. (1).
During the tth iteration jth variable of kth solution is update by XJ+1. Random number r1 and r2 have range between 0 to 1. During the tth iteration best candidate is Xjbest and worst candidate is X> orst.
Brief introduction of HEFT-TD and PEFT
HEFT-TD [6] is based on the duplication approach, this approach can be used to reduce the cost of communication between two dependent jobs. It is predicated on the concept that communicating across dependent activities running on the same system is completely free. In order to achieve this goal and reduce the cost overhead associated with inter-task communication, this strategy duplicates predecessor tasks.
The PEFT [7] is composed of two stages: the task prioritizing phase, which establishes task priorities, and the processor selection phase, which chooses the best processor for carrying out the current job. This algorithm forecasts by computing an Optimistic Cost Table while maintaining quadratic time complexity. The sum of a node earliest start time and computation time is called the Earliest Finish Time (EFT) of a node on a particular processor.
Computation of Execution Time and Cost
In this study, we focused not just on reducing execution costs while achieving the time constraint, but also managing load among various VMs. We took into account a VMs acquisition delay and performance fluctuation, both of which are crucial factors in reducing computation time in a diversified cloud computing environment.
ETVMk(Ti) :
Size г
Speed ум
(2)
Here ETVMk(T.) is task execution time which we obtained by dividing the size of the task (SizeT.) through the processing speed of the kth VM (SpeedVMf) as mentioned in eq. (2).
TTe.. = -
DataFileT
(3)
Here TTE.. is the time consumed to transfer the data
ij
between tasks scheduled in the different VMs. It is also known as communication time. It can be calculated by using the size of an output data file and average bandwidth P as mentioned in eq. (3) where Ej refers to the edge between parent task Tt to child task Tj, and DataFileT. is the output file of task T.
avail(VMk) = STT + { ET™№ , Tl 1(1 -PerVar))
(4)
avail(VMk) indicates when k<th VM is ready to execute new task. PerVar is performance variation of VM and STT. is the time which is estimated to start the execution as depicted in eq. (4).
STT. = acqdelay, if task is root node.
(5)
We used eq. (5) to compute the starting execution time of root node where acqdelay refers to booting time of VM, i.e. 60 sec.
STh = max{avail{VMk},maxTpEpred(T1){FTTp + TTepi}}.(6)
If the task is not a root node, then STTi is computed using eq. (6) where FTTi is the time which is estimated to finish the execution.
FTTj = STTj +
ЕТУМкЩ I (1 -PerVar))
(7)
By using the eq. (7) we computed the finishing time of task T.
If TET < D, TET = maxw{FT(Ti)}.
(8)
TET is total execution time which also included the termination delay of last VM which is executed till end of complition. TET is computed using eq. (8).
VMn
TEC =X Cy^VM^™-™-™™-^ (9)
-type(VMk)x
n=l [ Timelnterval
If TET does not violate the deadline constraint (D) then Total Execution Cost (TEC) is calculate as given in eq. (9) where Ciype(^VMk) denotes the cost of execution on VM of k type, while VMno ST and VMno ET is the start and end times of VM execution, respectively.
Our experiment takes into account three types of deadline constraints: hard, medium, and soft.
Deadline D = (a + 1) x minET(Wl).
(10)
Above, eq. (10) used to calculate the deadline where minET(^1) is the sum of time to start and execute all the tasks of workflow W. on the fastest VM, and a is a step length whose value is 0.4. The hard range is 0 to 1.2, the medium range is 1.2 to 2.8, and the soft range is 2.8 to 4.4. All above equations are taken from literature [22]. Proposed Algorithm
1. Determine the population size and termination criteria
2. Initialize one-one set of candidate solution using HEFT-TD and PEFT
3. Initialize remaining N-2 population using random technique with greedy approach
4. Compute the fitness of each candidate solutions in the population using fitness function using eq. (8) & eq. (9)
5. Find the best and worst candidate solutions
6. Apply the eq. (1) to all candidate solutions
7. Compute the Finish Time of candidate solution
8. According to eq. (10), check the deadline constraint of candidate solution
9. If any candidate solution does not satisfy deadline constraint according the eq. (8), then go to step 6 otherwise compute TEC using eq. (9)
10. If (j !)< jst) then
11 (Xj,best) = (Xj+ )
12. End If
13. Repeat steps 5 to 12 until termination criteria satisfied
14. Return best candidate solution
In our JAYA-based hybrid module, the population is initialized using three techniques. One individual is initialized using HEFT-TD, another one using the PEFT approach, and the remaining candidate solutions are generated using a random method with a greedy strategy. These approaches not only minimize the makespan but also reduce the computation cost. The fitness of each candidate is computed using eq. (8) and eq. (9), and the JAYA approach is applied using eq. (1) to find a more optimum solution. If the candidate solution satisfies the deadline constraint criteria, then the total execution cost is computed. These steps are iterated until we receive the optimum solution. Fig. 1 illustrates our model through flowchart to explain each step.
Evaluation of Performance Experimental Environment
We included various types of workflows as benchmarks like Montage, Cybershake, LIGO, and Epigenomics, with sizes of fifty, hundred, and five hundred tasks.
Montage represents an astronomy application where the majority of nodes focus more on I/O than processing.
The "Flexible Image Transport System" (FITS) format photographs used as input in this astronomy application are output as custom sky mosaics. Cybershake is employed by "Southern California Earthquake Center" to display earthquake risks in a particular area. It produces artificial seismograms. Additionally, this process requires a lot of memory and a central processing unit. Epigenomic is bioinformatics application workflow where the majority of the nodes are CPU-intensive. Changes in human cell gene function are referred to as epigenetic state and they are mapped on a genome-wide scale by an epigenomic approach. "Laser Interferometer Gravitational-wave Observatory" is acronym standing for LIGO. This scientific workflow uses a lot of CPU power and consumes a lot of memory. Numerous things happen in the universe that causes gravitational waves which it detects.
We executed the suggested model HJA in a JAVA-based robust environment and reached at a conclusion after performing each type of workflow 30 times. We examined five different types of VMs according to requirements [22] as illustrated in Table 1. We used the Amazon Elastic Block Store (EBS) average bandwidth of 20 kbps1. We employed
500 candidate solutions as size of population and a highest number of iterations of 250 in our experiment [23-26].
In the Fig. 1, i is each individual candidate solution from set of total number of N solutions.
Analysis of Experimental Result
Execution Cost and Makespan Analysis
Fig. 2 and 3 show our overall comparison of the baseline and our suggested HJA. The outcome of our experiment demonstrates the resilience of our suggested model HJA. HJA is 20.16 %, 16.58 %, 14.04 %, and 2.88 % less expensive than ABC, CEGA, CLGA, and CHGA respectively.
HJA has a 39.92 %, 9.05 %, 13.91 %, and 2.82 % shorter average makespan than ABC, CEGA, CLGA, and CHGA respectively.
Deadline and Load Balance Analysis
Table 2 illustrates the hit rate and Fig. 4 demonstrates balance of load among VMs.
If a processor hasn't any job, the load index measured value as zero, the load index increase according to the job assign to a processor.
1 Amazon Elastic Block Store. Available at: https://aws. amazon.com/ebs/ (accessed: 22.07.2020).
VMC, = PEnum x PEmips.
(11)
Table 1. Details of VMs in our experimental environment
VM Types Processing Capacity, GFLOPS ECUs (Cores) Memory, GB Disk, GB Cost /Hour, $
ml.Small 4.4 l(l) l.7 l60 0.04
ml.Large l7.6 4(2) 7.5 850 0.l6
ml.Xlarge 35.2 8(4) l5 l690 0.32
cl.Medium 22 5(2) l.7 350 0.20
cl.Xlarge 88 20(8) 7 l690 0.80
Table 2. Analysis of hit rate under deadline constraint, %
Deadline Algorithm Montage Cybershake LIGO Epigenomics
Hard CHGA 96.30 94.07 93.l0 92.30
ABC 78.l0 77.l0 79.l0 79.l2
HJA 95.62 93.00 92.56 90.l5
CEGA 92.34 88.48 88.50 83.49
CLGA 95.50 9l.48 9l.46 88.02
Crunch CHGA 99.90 99.80 99.74 99.83
ABC 82.08 80.35 8l.03 82.97
HJA 99.6l 99.82 99.60 99.8l
CEGA 99.50 99.62 99.50 99.6l
CLGA 99.5l 99.76 99.57 99.74
Soft CHGA 99.89 99.80 99.77 99.8l
ABC 99.8l 99.89 99.8l 99.90
HJA 83.45 82.00 85.78 86.57
CEGA 99.68 99.70 99.6l 99.50
CLGA 99.69 99.78 99.70 99.7l
Fig. 1. Flowchart of proposed model
We can calculate the capacity of a VMCf through product of the number of processing elements PEnum available in VMt and its execution speed in MIPS as mentioned in eq. (11). Here, PEnum indicates the number of processing elements in a particular VMt.
VMC = x vmc, i=l
(l2)
VMC indicates the execution capacity of all VMs which is equal to the sum of all VMCi as mentioned in eq. (12) where m is the total number of VMs.
L i =
JLTj
M
VMCi
(l3)
Load Lj is computed using eq. (13). Here, T refers to task length which is divided by VMC, and n is the number of total tasks. Task length is expressed as in MI (Million Instruction).
TL = X Li, i=l
(l4)
0.8 1.6 2.4 3.2 Deadline, min
0 0.8 1.6 2.4 3.2 Deadline, min
—CHGA — HJA ABC — CEGA CLGA
CHGA HJA ABC — CEGA CLGA
«л 60
о U
1.6 2.4 3.2 Deadline, min
0 0.8 1.6 2.4 3.2 Deadline, min
Fig. 2. Analysis of Cost: Montage Workflow (a); Cybershake Workflow (b); LIGO Workflow (c); Epigenomics Workflow (d)
Total Load (TL) represents the load on all VMs in a given data center which is calculated as given in eq. (14), where i refers to the number of VMs till m.
LCpu =
TL VMC
(15)
Load capacity per unit is computed as given in eq. (15).
TH, = LCpu x VMC. (16)
The threshold value THi of any VMi is computed as mentioned in eq. (16).
For any VM, if the total load on VMt is less than its threshold value THt then it is under-loaded, if it is equal to its THi then it is balanced, and if it is greater than its THi then it is over-loaded. In our experiment, we examine the variation of load on 5 VMs under the baseline algorithms as well as our proposed HJA. If load is greater than 100 percentages that means machine is overloaded; if less than 100 percentages, i.e., it is under-loaded.
Fig. 4 illustrates how our HJA outperforms competing baseline techniques in terms of load balancing across all 5 VMs. VMX has the highest load of 112, and VM3 has the lowest load of 95. It indicates that VM3 is under loaded by
-5 whereas VMl is overloaded by +12. This demonstrates the sturdiness of the HJA model we've suggested.
Discussion
Three strategies are used to initialize the population in our hybrid module based on JAYA. The remaining potential solutions are created using a random method with a greedy strategy, and one individual is initialized using HEFT-TD, another using the PEFT methodology. These methods lower the cost of calculation while simultaneously minimizing the makespan. To arrive at a more ideal option, the fitness of each candidate is calculated and the JAYA technique is used. The overall execution cost is calculated if the candidate solution meets the deadline constraint requirements. We repeat these procedures until we find the optimum solution. The results of our experiment show how robust our proposed model HJA is. ABC, CEGA, CLGA, and CHGA are each 20.16 %, 16.58 %, 14.04 %, and 2.88 % more expensive than HJA. In comparison to ABC, CEGA, CLGA, and CHGA, respectively, HJA has shorter average makespans of 39.92 %, 9.05 %, 13.9131 %, and 2.82 %. In terms of load balancing across all 5 VMs, our HJA performs better than competing baseline
—CHGA — HJA ABC — CEGA CLGA
— CHGA —HJA ABC —CEGA CLGA
■Ü
1.6 2.4 3.2 Deadline, min
0 0.8 1.6 2.4 3.2 4 Deadline, min
—CHGA —HJA ABC —CEGA CLGA
1100
■Э
900
500
—CHGA — HJA ABC —CEGA CLGA
3500
2500
1 1500
500
1.6 2.4 3.2 4 0 0.8 1.6 2.4 3.2
Deadline, min Deadline, min
Fig. 3. Analysis of Makespan: Montage Workflow (a); Cybershake Workflow (b); LIGO Workflow (c); Epigenomics Workflow (d)
methodologies. The load on VM1 is the highest at 112, and the load on VM3 is the lowest at 95. It shows that VM1 is overloaded by +12, whereas VM3 is under-loaded by -5. This illustrates how resilient the HJA paradigm we've proposed is.
CLGA
VM1 VM2 VM3 VM4 Types of Virtual Machine
VM5
Conclusion and Future Direction
We revealed our meta-heuristic, cost-effective, load-balanced hybrid evolutionary strategy to schedule scientific process. We applied the HEFT-TD and PEFT approaches to minimize the execution cost of workflows and to balance the load across VMs where configuration of VMs is varying. We extensively evaluated four types of workflows which are belong to scientific domains with varying task sizes within a user-defined deadline, taking into account three parameters: makespan, computing cost, and load balance. Our experimental findings have shown that the suggested HJA algorithm outperforms the ABC, CEGA, CLGA, and CHGA in terms of computational cost and execution time as well as load balancing across virtual machines.
In the future, we would like to address the dynamic nature of workflow with varying nature of communication bandwidth and communication delays throughout data centers using the most recent metaheuristics strategies along with machine learning.
Fig. 4. Balance of load among various VMs
References
1. Yu J., Buyya R. A taxonomy of scientific workflow systems for grid computing. ACM SIGMOD Record, 2005, vol. 34, no. 3. pp. 44-49. https://doi.org/10.1145/1084805.1084814
2. Singhal S., Grover J. Hybrid biogeography algorithm for reducing power consumption in cloud computing. Proc. of the 2017 International Conference on Advances in Computing, Communications and Informatics (ICACCI), 2017, pp. 121-124. https://doi.org/10.1109/ICACCI.2017.8125827
3. Bothra S.K., Singhal S. Nature-inspired metaheuristic scheduling algorithms in cloud: A systematic review. Scientific and Technical Journal of Information Technologies, Mechanics and Optics, 2021, vol. 21, no. 4, pp. 463-472. https://doi.org/10.17586/2226-1494-2021-21-4-463-472
4. Bhatt C., Singhal S. Anatomy of virtual machine placement techniques in cloud. Lecture Notes in Networks and Systems, 2022, vol. 373, pp. 609-626. https://doi.org/10.1007/978-981-16-8721-1_59
5. Singhal S., Patel J. Load balancing scheduling algorithm for concurrent workflow. Computing and Informatics, 2018, vol. 37, no. 2, pp. 311-326. https://doi.org/10.4149/cai_2018_2_311
6. NoorianTalouki R., Hosseini Shirvani M., Motameni H. A heuristic-based task scheduling algorithm for scientific workflows in heterogeneous cloud computing platforms. Journal of King Saud University — Computer and Information Sciences, 2022, vol. 34, no. 8, pp. 4902-4913. https://doi.org/10.1016/jjksuci.2021.05.011
7. Arabnejad H., Barbosa J.G. List scheduling algorithm for heterogeneous systems by an optimistic cost table. IEEE Transactions on Parallel and Distributed Systems, 2014, vol. 25, no. 3, pp. 682694. https://doi.org/10.1109/tpds.2013.57
8. Shubham, Gupta R., Gajera V., Jana P.K. An effective multi-objective workflow scheduling in cloud computing: a pso based approach. Proc. of the 2016 Ninth International Conference on Contemporary Computing (IC3), 2016, pp. 1-6. https://doi.org/10.1109/ ic3.2016.7880196
9. Lal A., Krishna C.R. Critical path-based ant colony optimization for scientific workflow scheduling in cloud computing under deadline constraint. Advances in Intelligent Systems and Computing, 2018, vol. 696, pp. 447-461. https://doi.org/10.1007/978-981-10-7386-1_39
10. Xu R., Wang Y., Cheng Y., Zhu Y., Xie Y., Sani A.S., Yuan D. Improved particle swarm optimization based workflow scheduling in cloud-fog environment. Lecture Notes in Business Information Processing, 2019, vol. 342, pp. 337-347. https://doi.org/10.1007/978-3-030-11641-5_27
11. Meshkati J., Safi-Esfahani F. Energy-aware resource utilization based on particle swarm optimization and artificial bee colony algorithms in cloud computing. The Journal of Supercomputing, 2019, vol. 75, no. 5, pp. 2455-2496. https://doi.org/10.1007/s11227-018-2626-9
12. Mohanty S., Patra P.K., Ray M., Mohapatra S. An approach for load balancing in cloud computing using JAYA algorithm. International Journal of Information Technology and Web Engineering, 2019, vol. 14, no. 1, pp. 27-41. https://doi.org/10.4018/IJITWE.2019010102
13. Nayak S.C., Parida S., Tripathy C., Pattnaik P.K. Dynamic backfilling algorithm to increase resource utilization in cloud computing. International Journal of Information Technology and Web Engineering, 2019, vol. 14, no. 1, pp. 1-26. https://doi.org/10.4018/ IJITWE.2019010101
14. Masadeh R.M.T., Sharieh A.-A.A., Mahafzah B.A. Humpback whale optimization algorithm based on vocal behavior for task scheduling in cloud computing. International Journal of Advanced Science and Technology, 2019, vol. 13, no. 3, pp. 121-140.
15. Sanaj M.S., Prathap P.J. An efficient approach to the map-reduce framework and genetic algorithm based whale optimization algorithm for task scheduling in cloud computing environment. Materials Today: Proceedings, 2021, vol. 37, pp. 3199-3208. https://doi. org/10.1016/j.matpr.2020.09.064
16. Gu Y., Budati C. Energy-aware workflow scheduling and optimization in clouds using bat algorithm. Future Generation Computer Systems, 2020, vol. 113, pp. 106-112. https://doi.org/10.1016/j. future.2020.06.031
17. Ullah A., Nawi N.M., Khan M.H. BAT algorithm used for load balancing purpose in cloud computing: an overview. International Journal of High Performance Computing and Networking, 2020, vol. 16, no. 1, pp. 43. https://doi.org/10.1504/ijhpcn.2020.110258
Литература
1. Yu J., Buyya R. A taxonomy of scientific workflow systems for grid computing // ACM SIGMOD Record. 2005. V. 34. N 3. P. 44-49. https://doi.org/10.1145/1084805.1084814
2. Singhal S., Grover J. Hybrid biogeography algorithm for reducing power consumption in cloud computing // Proc. of the 2017 International Conference on Advances in Computing, Communications and Informatics (ICACCI). 2017. P. 121-124. https://doi.org/10.1109/ICACCI.2017.8125827
3. Bothra S.K., Singhal S. Nature-inspired metaheuristic scheduling algorithms in cloud: A systematic review // Научно-технический вестник информационных технологий, механики и оптики. 2021. V. 21. N 4. P. 463-472. https://doi.org/10.17586/2226-1494-2021-21-4-463-472
4. Bhatt C., Singhal S. Anatomy of virtual machine placement techniques in cloud // Lecture Notes in Networks and Systems. 2022. V. 373. P. 609-626. https://doi.org/10.1007/978-981-16-8721-1_59
5. Singhal S., Patel J. Load balancing scheduling algorithm for concurrent workflow // Computing and Informatics. 2018. V. 37. N 2. P. 311-326. https://doi.org/10.4149/cai_2018_2_311
6. NoorianTalouki R., Hosseini Shirvani M., Motameni H. A heuristic-based task scheduling algorithm for scientific workflows in heterogeneous cloud computing platforms // Journal of King Saud University - Computer and Information Sciences. 2022. V. 34. N 8. P. 4902-4913. https://doi.org/10.1016/j.jksuci.2021.05.011
7. Arabnejad H., Barbosa J.G. List scheduling algorithm for heterogeneous systems by an optimistic cost table // IEEE Transactions on Parallel and Distributed Systems. 2014. V. 25. N 3. P. 682-694. https://doi.org/10.1109/tpds.2013.57
8. Shubham, Gupta R., Gajera V., Jana P.K. An effective multi-objective workflow scheduling in cloud computing: a pso based approach // Proc. of the 2016 Ninth International Conference on Contemporary Computing (IC3). 2016. P. 1-6. https://doi.org/10.1109/ ic3.2016.7880196
9. Lal A., Krishna C.R. Critical path-based ant colony optimization for scientific workflow scheduling in cloud computing under deadline constraint // Advances in Intelligent Systems and Computing. 2018. V. 696. P. 447-461. https://doi.org/10.1007/978-981-10-7386-1_39
10. Xu R., Wang Y., Cheng Y., Zhu Y., Xie Y., Sani A.S., Yuan D. Improved particle swarm optimization based workflow scheduling in cloud-fog environment // Lecture Notes in Business Information Processing. 2019. V. 342. P. 337-347. https://doi.org/10.1007/978-3-030-11641-5_27
11. Meshkati J., Safi-Esfahani F. Energy-aware resource utilization based on particle swarm optimization and artificial bee colony algorithms in cloud computing // The Journal of Supercomputing. 2019. V. 75. N 5. P. 2455-2496. https://doi.org/10.1007/s11227-018-2626-9
12. Mohanty S., Patra P.K., Ray M., Mohapatra S. An approach for load balancing in cloud computing using JAYA algorithm // International Journal of Information Technology and Web Engineering. 2019. V. 14. N 1. P. 27-41. https://doi.org/10.4018/IJITWE.2019010102
13. Nayak S.C., Parida S., Tripathy C., Pattnaik P.K. Dynamic backfilling algorithm to increase resource utilization in cloud computing // International Journal of Information Technology and Web Engineering. 2019. V. 14. N 1. P. 1-26. https://doi.org/10.4018/ IJITWE.2019010101
14. Masadeh R.M.T., Sharieh A.-A.A., Mahafzah B.A. Humpback whale optimization algorithm based on vocal behavior for task scheduling in cloud computing // International Journal of Advanced Science and Technology. 2019. V. 13. N 3. P. 121-140.
15. Sanaj M.S., Prathap P.J. An efficient approach to the map-reduce framework and genetic algorithm based whale optimization algorithm for task scheduling in cloud computing environment // Materials Today: Proceedings. 2021. V. 37. P. 3199-3208. https://doi. org/10.1016/j.matpr.2020.09.064
16. Gu Y., Budati C. Energy-aware workflow scheduling and optimization in clouds using bat algorithm // Future Generation Computer Systems. 2020. V. 113. P. 106-112. https://doi.org/10.1016/j.future.2020.06.031
17. Ullah A., Nawi N.M., Khan M.H. BAT algorithm used for load balancing purpose in cloud computing: an overview // International Journal of High Performance Computing and Networking. 2020. V. 16. N 1. P. 43. https://doi.org/10.1504/ijhpcn.2020.110258
18. Aziza H., Krichen S. A hybrid genetic algorithm for scientific workflow scheduling in cloud environment // Neural Computing and
18. Aziza H., Krichen S. A hybrid genetic algorithm for scientific workflow scheduling in cloud environment. Neural Computing and Applications, 2020, vol. 32, no. 18, pp. 15263-15278. https://doi. org/10.1007/s00521-020-04878-8
19. Sardaraz M., Tahir M. A parallel multi-objective genetic algorithm for scheduling scientific workflows in cloud computing. International Journal of Distributed Sensor Networks, 2020, vol. 16, no. 8. https:// doi.org/10.1177/1550147720949142
20. Gupta S., Agarwal I., Singh R.S. Workflow scheduling using Jaya algorithm in cloud. Concurrency and Computation: Practice and Experience, 2019, vol. 31, no. 17, pp. e5251. https://doi.org/10.1002/ cpe.5251
21. Rao R.V. Jaya: A simple and new optimization algorithm for solving constrained and unconstrained optimization problems. International Journal of Industrial Engineering Computations, 2016, vol. 7, no. 1, pp. 19-34. https://doi.org/10.5267/j.ijiec.2015.8.004
22. Bothra S.K., Singhal S., Goyal H. Deadline-constrained cost-effective load-balanced improved genetic algorithm for workflow scheduling. International Journal of Information Technology and Web Engineering, 2021, vol. 16, no. 4, pp. 1-34. https://doi.org/10.4018/ IJITWE.2021100101
23. Bothra S.K., Singhal S., Goyal H. Cost effective hybrid genetic algorithm for workflow scheduling in cloud. System Research and Information Technologies, 2022, vol. 3, pp. 121-138. https://doi.
24.
25.
26.
Applications. 2020. V. 32. N l8. P. l5263-l5278. https://doi. org/l0.l007/s0052l-020-04878-8
19. Sardaraz M., Tahir M. A parallel multi-objective genetic algorithm for scheduling scientific workflows in cloud computing // International Journal of Distributed Sensor Networks. 2020. V. l6. N 8. https://doi. org/l0.ll77/l550l47720949l42
20. Gupta S., Agarwal I., Singh R.S. Workflow scheduling using Jaya algorithm in cloud // Concurrency and Computation: Practice and Experience. 20l9. V. 3l. N l7. P. e525l. https://doi.org/l0.l002/ cpe.525l
21. Rao R.V. Jaya: A simple and new optimization algorithm for solving constrained and unconstrained optimization problems // International Journal of Industrial Engineering Computations. 20l6. V. 7. N l. P. l9-34. https://doi.org/l0.5267/j.ijiec.20l5.8.004
22. Bothra S.K., Singhal S., Goyal H. Deadline-constrained cost-effective load-balanced improved genetic algorithm for workflow scheduling // International Journal of Information Technology and Web Engineering. 202l. V. l6. N 4. P. l-34. https://doi.org/l0.40l8/ IJITWE.202ll00l0l
23. Bothra S.K., Singhal S., Goyal H. Cost effective hybrid genetic algorithm for workflow scheduling in cloud // System Research and Information Technologies. 2022. V. 3. P. l2l-l38. https://doi. org/l0.20535/srit.2308-8893.2022.3.08
24. Голосов П.Е., Гостев И.М. Имитационная модель облачных вы-
org/10.20535/srit.2308-8893.2022.3.08
Golosov P.E., Gostev I.M. Cloud computing simulation model with a sporadic mechanism of parallel task solving control. Scientific and Technical Journal of Information Technologies, Mechanics and Optics, 2022, vol. 22, no. 2, pp. 269-278. (in Russian). https://doi. org/10.17586/2226-1494-2022-22-2-269-278
Arkhipov I.V. Genetic algorithm application for multi-criteria scheduling problem. Scientific and Technical Journal of Information Technologies, Mechanics and Optics, 2015, vol. 15, no. 3, pp. 525531. (in Russian). https://doi.org/10.17586/2226-1494-2015-15-3-525-531
Becker M., Gatchin Y.A., Karmanovskiy N.S., Terentiev A.O., Fyodorov D.Y. Information security in cloud computing: problems and prospects. Scientific and Technical Journal of Information Technologies, Mechanics and Optics, 2011, vol. 11, no. 1, pp. 97-102. (in Russian)
числений со спорадическим механизмом управления параллельным решением задач // Научно-технический вестник информационных технологий, механики и оптики. 2022. Т. 22. № 2. С. 269-278. https://doi.org/10.17586/2226-1494-2022-22-2-269-278
25. Архипов И.В. Применение генетического алгоритма для многокритериальной задачи календарного планирования // Научно-технический вестник информационных технологий, механики и оптики. 2015. Т. 15. № 3. С. 525-531. https://doi.org/10.17586/2226-1494-2015-15-3-525-531
26. Беккер М.Я., Гатчин Ю.А., Кармановский Н.С., Терентьев А.О., Федоров Д.Ю. Информационная безопасность при облачных вычислениях: проблемы и перспективы // Научно-технический вестник Санкт-Петербургского государственного университета информационных технологий, механики и оптики. 2011. Т. 11. № 1. С. 97-102.
Authors
Sandeep K. Bothra — Researcher, Manipal University Jaipur, Jaipur, Rajasthan, 303007, India, S3 57279416100, https://orcid.org/0000-0003-0555-569X, bothrajain@gmail.com
Sunita Singhal — PhD, Associate Professor, Manipal University Jaipur, Jaipur, Rajasthan, 303007, India, S3 57194063002, https://orcid.org/0000-0003-2462-8102, sunita.singhal@jaipur.manipal.edu
Hemlata Goyal — PhD, Assistant Professor, Manipal University Jaipur, Jaipur, Rajasthan, 303007, India, S3 57202709665, https://orcid.org/0000-0003-1344-0921, hemlata.goyal@jaipur.manipal.edu
Авторы
Ботра Сандал Кумар — исследователь, Манипальский университет Джайпура, Джайпур, Раджастхан, 303007, Индия, 57279416100, https://orcid.org/0000-0003-0555-569X, bothrajain@gmail.com Сингхал Сунита — PhD, доцент, Манипальский университет Джайпура, Джайпур, Раджастхан, 303007, Индия, 57194063002, https://orcid.org/0000-0003-2462-8102, sunita.singhal@jaipur.manipal. edu
Гоял Хемлата — PhD, доцент, Манипальский университет Джайпура, Джайпур, Раджастхан, 303007, Индия, 57202709665, https://orcid.org/0000-0003-1344-0921, hemlata.goyal@jaipur.manipal. edu
Received 25.07.2022
Approved after reviewing 11.01.2023
Accepted 14.03.2023
Статья поступила в редакцию 25.07.2022 Одобрена после рецензирования 11.01.2023 Принята к печати 14.03.2023
Работа доступна по лицензии Creative Commons «Attribution-NonCommercial»
