Hussain Altaf, Aleem Muhammad, Ur Rehman Atiq, Arshad Umer
Department of Computer Science, KICSIT Campus, Institute of Space Technology, Islamabad, Pakistan.
School of Computing, Mathematics and Data Science, Coventry University, London, United Kingdom.
PeerJ Comput Sci. 2025 Mar 18;11:e2739. doi: 10.7717/peerj-cs.2739. eCollection 2025.
Cloud computing provides an opportunity to gain access to the large-scale and high-speed resources without establishing your own computing infrastructure for executing the high-performance computing (HPC) applications. Cloud has the computing resources (., computation power, storage, operating system, network, and database .) as a public utility and provides services to the end users on a pay-as-you-go model. From past several years, the efficient utilization of resources on a compute cloud has become a prime interest for the scientific community. One of the key reasons behind inefficient resource utilization is the imbalance distribution of workload while executing the HPC applications in a heterogenous computing environment. The static scheduling technique usually produces lower resource utilization and higher makespan, while the dynamic scheduling achieves better resource utilization and load-balancing by incorporating a dynamic resource pool. The dynamic techniques lead to increased overhead by requiring a continuous system monitoring, job requirement assessments and real-time allocation decisions. This additional load has the potential to impact the performance and responsiveness on computing system. In this article, a dynamic enhanced resource-aware load balancing algorithm (DE-RALBA) is proposed to mitigate the load-imbalance in job scheduling by considering the computing capabilities of all VMs in cloud computing. The empirical assessments are performed on CloudSim simulator using instances of two scientific benchmark datasets (., heterogeneous computing scheduling problems (HCSP) instances and Google Cloud Jobs (GoCJ) dataset). The obtained results revealed that the DE-RALBA mitigates the load imbalance and provides a significant improvement in terms of makespan and resource utilization against existing algorithms, namely PSSLB, PSSELB, Dynamic MaxMin, and DRALBA. Using HCSP instances, the DE-RALBA algorithm achieves up to 52.35% improved resources utilization as compared to existing technique, while more superior resource utilization is achieved using the GoCJ dataset.
云计算提供了一个机会，无需建立自己的计算基础设施来执行高性能计算（HPC）应用程序，就能访问大规模和高速资源。云将计算资源（如计算能力、存储、操作系统、网络和数据库等）作为一种公用事业，并以按需付费模式向最终用户提供服务。在过去几年中，计算云资源的高效利用已成为科学界的主要关注点。资源利用效率低下的一个关键原因是在异构计算环境中执行HPC应用程序时工作负载分布不均衡。静态调度技术通常会导致较低的资源利用率和较长的完工时间，而动态调度通过引入动态资源池实现了更好的资源利用和负载均衡。动态技术由于需要持续的系统监控、作业需求评估和实时分配决策，导致开销增加。这种额外的负载有可能影响计算系统的性能和响应能力。在本文中，提出了一种动态增强资源感知负载均衡算法（DE-RALBA），通过考虑云计算中所有虚拟机的计算能力来减轻作业调度中的负载不平衡。使用两个科学基准数据集（异构计算调度问题（HCSP）实例和谷歌云作业（GoCJ）数据集）的实例在CloudSim模拟器上进行了实证评估。获得的结果表明，与现有算法（即PSSLB、PSSELB、动态最大最小算法和DRALBA）相比，DE-RALBA减轻了负载不平衡，并在完工时间和资源利用方面有显著改进。使用HCSP实例时，与现有技术相比，DE-RALBA算法的资源利用率提高了52.35%，而使用GoCJ数据集时实现了更优的资源利用率。
