Research Articles Open Access

An Enhanced Hybrid PSO-LSTM Framework for Energy-Aware Task Scheduling in Heterogeneous Cloud Computing Environments

Zain ul Huda
Available online January 1, 2025
PDF

Abstract

Cloud task scheduling is an NP-hard optimization problem in which submitted tasks must be mapped to virtual machines under competing performance objectives. Particle Swarm Optimization (PSO) is widely used for this problem because of its simple update procedure, but the standard algorithm can converge prematurely. In the selected PSO scheduling papers, task sets are also fixed before optimization and next-window workload prediction is not incorporated into the particle update. This paper presents a Hybrid PSO-LSTM framework for energy-aware scheduling in heterogeneous cloud environments. A Long Short-Term Memory (LSTM) model learns historical task-arrival sequences and estimates the workload for the next scheduling window. The prediction stage provides an additional guidance term for PSO, while candidate task-to-VM assignments are evaluated using makespan, energy consumption, and resource utilization. The framework is implemented in CloudSim and compared with standard PSO, Genetic Algorithm, Round Robin, First-Come, First-Served, and Whale Optimization Algorithm using workloads of up to 1000 tasks and 50 virtual machines. Across the reported comparisons, PSO-LSTM records lower makespan and energy consumption and higher resource utilization and throughput. The makespan-based scalability experiment shows the same relative trend within the simulated workload range.

Keywords

Cloud computing Task scheduling Particle Swarm Optimization Long Short-Term Memory Energy-aware scheduling

Introduction

Cloud computing provides virtualized processing, storage, and networking resources that can be provisioned according to demand [19], [20]. In this environment, task scheduling determines how submitted jobs are assigned to the available virtual machines (VMs). The assignment affects completion time, resource utilization, and the operating cost of the cloud platform [2], [6], [17].

Cloud task scheduling is commonly modeled as an optimization problem in which a set of tasks must be mapped to a set of heterogeneous VMs. The problem is NP-hard, so exhaustive search becomes impractical as the number of possible assignments grows [6]. Metaheuristic methods, including Particle Swarm Optimization (PSO), are therefore widely used to search for feasible schedules within reasonable computation time [2], [9].

PSO has a simple update rule, but the standard algorithm can lose population diversity and converge prematurely [9]. In addition, several PSO-based cloud schedulers are evaluated on task sets that are fixed before optimization begins [1], [2], [6]. Such formulations do not use an estimate of the workload expected in the next scheduling window. Energy is also not included in every PSO scheduling objective [1]-[4], although energy-aware PSO schedulers have been reported [6]. Multi-objective scheduling is likewise well established in the literature [6]-[8], [17]. The distinction pursued in this paper is therefore not multi-objective scheduling alone; it is the use of workload prediction as an additional PSO guidance signal while evaluating makespan, energy consumption, and resource utilization together.

This paper presents a Hybrid PSO-LSTM framework for energy-aware task scheduling in a heterogeneous cloud environment. The LSTM component learns historical task-arrival patterns and predicts the workload for the next scheduling window. The enhanced PSO component uses prediction-derived guidance during particle updates. Candidate task-to-VM assignments are evaluated using makespan, energy consumption, and resource utilization.

The main contributions are as follows:

  • We couple LSTM-based workload prediction with PSO-based task scheduling so that anticipated workload conditions can influence the particle search.

  • We introduce a PSO guidance term derived from the workload-prediction stage and compare the resulting scheduler with standard PSO and four additional baselines.

  • The scheduling objective evaluates makespan, energy consumption, and resource utilization together. This combination differs from the objectives used by the selected comparison papers, although multi-objective cloud scheduling itself is not new.

  • The framework is implemented in CloudSim and evaluated with workloads of 100, 200, 500, and 1000 tasks over 10, 20, 30, and 50 VMs.

  • The reported experiments include makespan, energy consumption, cost, throughput, convergence, response time, load distribution, memory use, and scalability measures.

The remainder of this paper is organized as follows. Section 2 reviews the related work. Section 3 presents the proposed Hybrid PSO-LSTM framework. Section 4 examines the selected base PSO paper and identifies the limitations that are relevant to the proposed framework. Section 5 shows the results. Section 6 concludes the paper with the future work.

Complete Article

The complete article, including all figures, tables, equations and algorithms, is available in the official publication PDF.

Conclusion

In this paper, a Hybrid PSO-LSTM scheduler is developed for heterogeneous cloud environments. Historical task arrivals are processed by the LSTM to estimate the workload of the next scheduling window. That estimate enters the PSO velocity update as an additional guidance term. Makespan, energy consumption, and resource utilization are then evaluated together in the fitness function. The design is intended to address the static-workload assumption, limited energy awareness, single-objective formulation, and premature-convergence issue discussed earlier. The implementation is evaluated in CloudSim against standard PSO, GA, RR, FCFS, and WOA with workloads of up to 1000 tasks and 50 VMs. For the simulated cases, PSO-LSTM records lower makespan, energy consumption, cost, and response time, together with higher resource utilization, throughput, scheduling success rate, and more even VM utilization than the compared baselines. These results are limited to the reported simulation settings.

Future work will test larger production-scale deployments and workloads with more dynamic and unpredictable task-arrival patterns.

References

  1. S. H. Anbarkhan and M. A. Rakrouki, “An Enhanced PSO Algorithm for Scheduling Workflow Tasks in Cloud Computing, ” Electronics, vol. 12, no. 12, Art. no. 2580, 2023, doi: 10.3390/electronics12122580
  2. S. Nabi, M. Ahmad, M. Ibrahim, and H. Hamam, “AdPSO: Adaptive PSO-Based Task Scheduling Approach for Cloud Computing, ” Sensors, vol. 22, no. 3, Art. no. 920, 2022, doi: 10.3390/s22030920
  3. Y. Wang and X. Q. Zuo, “An Effective Cloud Workflow Scheduling Approach Combining PSO and Idle Time Slot-Aware Rules, ” IEEE/CAA J. Autom. Sinica, vol. 8, no. 5, pp. 1079-1094, May 2021, doi: 10.1109/JAS.2021.1003982
  4. A. R. Shaheen and S. Santhosh Kumar, “Tasks Scheduling in Cloud Environment Using PSO-BATS with MLRHE, ” Intell. Automat. Soft Comput., vol. 35, no. 3, pp. 2963-2978, 2023, doi: 10.32604/iasc.2023.025780
  5. G. Rjoub, J. Bentahar, O. A. Wahab, and A. S. Bataineh, “Deep and Reinforcement Learning for Automated Task Scheduling in Large-Scale Cloud Computing Systems, ” Concurr. Comput. Pract. Exp., vol. 33, no. 23, Art. no. e5919, 2021, doi: 10.1002/cpe.5919
  6. S. A. Alsaidy, A. D. Abbood, and M. A. Sahib, “Heuristic Initialization of PSO Task Scheduling Algorithm in Cloud Computing, ” J. King Saud Univ. Comput. Inf. Sci., vol. 34, no. 6, Part A, pp. 2370-2382, 2022, doi: 10.1016/j.jksuci.2020.11.002
  7. A. Y. Hamed and M. H. Alkinani, “Task Scheduling Optimization in Cloud Computing Based on Genetic Algorithms, ” Comput. Mater. Contin., vol. 69, no. 3, pp. 3289-3301, 2021, doi: 10.32604/cmc.2021.018658
  8. L. W. Jia, K. Li, and X. Shi, “Cloud Computing Task Scheduling Model Based on Improved Whale Optimization Algorithm, ” Wirel. Commun. Mob. Comput., vol. 2021, Art. no. 4888154, 2021, doi: 10.1155/2021/4888154
  9. T. M. Shami et al., “Particle Swarm Optimization: A Comprehensive Survey, ” IEEE Access, vol. 10, pp. 10031-10061, 2022, doi: 10.1109/ACCESS.2022.3142859
  10. J. Bi, H. Ma, H. Yuan, and J. Zhang, “Accurate Prediction of Workloads and Resources with Multi-Head Attention and Hybrid LSTM for Cloud Data Centers, ” IEEE Trans. Sustain. Comput., vol. 8, no. 3, pp. 375-384, 2023, doi: 10.1109/TSUSC.2023.3259522
  11. Y. Xing, “Work Scheduling in Cloud Network Based on Deep Q-LSTM Models for Efficient Resource Utilization, ” J. Grid Comput., vol. 22, Art. no. 36, 2024, doi: 10.1007/s10723-024-09746-6
  12. M. T. Islam, S. Karunasekera, and R. Buyya, “Performance and Cost-Efficient Spark Job Scheduling Based on Deep Reinforcement Learning in Cloud Computing Environments, ” IEEE Trans. Parallel Distrib. Syst., vol. 33, no. 7, pp. 1695-1710, 2022, doi: 10.1109/TPDS.2021.3124670
  13. A. Katal, S. Dahiya, and T. Choudhury, “Energy Efficiency in Cloud Computing Data Centers: A Survey on Software Technologies, ” Clust. Comput., vol. 26, no. 3, pp. 1845-1875, 2023, doi: 10.1007/s10586-022-03713-0
  14. B. Wang, C. Wang, W. Huang, Y. Song, and X. Qin, “Security-Aware Task Scheduling with Deadline Constraints on Heterogeneous Hybrid Clouds, ” J. Parallel Distrib. Comput., vol. 153, pp. 15-28, 2021, doi: 10.1016/j.jpdc.2021.03.003
  15. S. Ouhame, Y. Hadi, and A. Ullah, “An Efficient Forecasting Approach for Resource Utilization in Cloud Data Center Using CNN-LSTM Model, ” Neural Comput. Appl., vol. 33, no. 16, pp. 10043-10055, 2021, doi: 10.1007/s00521-021-05770-9
  16. J. Dogani, F. Khunjush, M. R. Mahmoudi, and M. Seydali, “Multivariate Workload and Resource Prediction in Cloud Computing Using CNN and GRU by Attention Mechanism, ” J. Supercomput., vol. 79, no. 3, pp. 3437-3470, 2023, doi: 10.1007/s11227-022-04782-z
  17. A. Moazeni, R. Khorsand, and M. Ramezanpour, “Dynamic Resource Allocation Using an Adaptive Multi-Objective Teaching-Learning Based Optimization Algorithm in Cloud, ” IEEE Access, vol. 11, pp. 23407-23419, 2023, doi: 10.1109/ACCESS.2023.3247639
  18. Y. Alahmad, T. Daradkeh, and A. Agarwal, “Proactive Failure-Aware Task Scheduling Framework for Cloud Computing, ” IEEE Access, vol. 9, pp. 106152-106168, 2021, doi: 10.1109/ACCESS.2021.3101147
  19. T. Khan, W. Tian, G. Zhou, S. Ilager, M. Gong, and R. Buyya, “Machine Learning (ML)-Centric Resource Management in Cloud Computing: A Review and Future Directions, ” J. Netw. Comput. Appl., vol. 204, Art. no. 103405, 2022, doi: 10.1016/j.jnca.2022.103405
  20. 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. Exp., vol. 41, no. 1, pp. 23-50, 2011, doi: 10.1002/spe.995
37 31

Similar Articles

You may also start an advanced similarity search for this article.