Year: 2025
Type: article
Abstract: Cloud computing has emerged as a promising solution for meeting the computational objectives of high-performance systems by efficiently scheduling cloud workloads to available resources. However, it becomes problematic to achieve multiple objectives simultaneously, such as optimizing throughput, minimizing makespan, and improving resource utilization for effective Task Scheduling (TS). In response to this issue, the TS, based on many optimization algorithms, was recently developed. These algorithms optimally allocate tasks to appropriate resources to achieve multiple objectives simultaneously. Most of these algorithms enhance the resource usage and the network throughput while minimizing the makespan, but ignore memory and bandwidth consumption. Combining plan-based and backfilling strategies, this article proposes a hybrid TS algorithm that could fulfill numerous criteria for effective TS. Initially, the reliability of the job runtimes is pre-dicted using a meta-learner called a Gated Recurrent Unit (GRU) network. Then, the tasks are sorted into two categories: reliable and unreli-able. The Dove Optimization Algorithm (DOA) is used for plan-based scheduling of jobs with higher runtime estimates, while the Analytic Hierarchy Process (AHP) method is used to backfill jobs with lower estimated runtimes. In addition, a dynamic allocation of CPU cores is made for backfilling based on the resource limit ratio of reliable jobs to newly requested jobs. Lastly, the simulation results show that the proposed algorithm outperforms the existing TS algorithms. GRU-DOA-AHP reduces makespan by 20.38%, increases throughput by 30%, and decreases memory usage by up to 26.19% compared to traditional algorithms.
DOI: 10.14419/njb1me19
Authors: B. Suganya, Dr. R. Padmapriya
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4416827430
FWCI:
Cites:
Cited by:
Related to: