A Novel genetic multi objective task scheduling algorithm in cloud computing

Abstract

Cloud computing is a trending cutting-edge model that allows users to utilize services through the internet. It offers on-demand scalable resources and compensates customers based on their usage. The cloud platform provides extensive computing power with a variety of configurable options. Effective resource management is crucial for optimizing the underlying system and ensuring that both cloud providers and users meet the Quality-of-Service (QoS) requirements specified in Service Level Agreements (SLAs). Numerous strategies are employed for cloud resource management, with scheduling being a prominent approach for optimizing cloud systems. Task and workflow scheduling are key unresolved challenges within cloud computing scheduling. Assigning tasks to suitable cloud resources is known to be an NP-complete problem. There are various factors affecting the execution of different workloads. Hence there is no single optimal solution under a given polynomial time. Many factors and objectives need to be considered when trying to find an optimal schedule. The total execution time of virtual machines has not been considered by previous studies as a scheduling parameter in the quest for an optimal schedule using meta-heuristic methods such as genetic algorithms. This study introduces G-TRETA, a new multi-objective genetic algorithm that aims to find an optimal schedule considering both total resource execution time and cost. The proposed algorithm was compared with FCFS, MaxMin, MinMin and MCT algorithms for the degree of imbalance, makespan, cost, and system throughput. CloudSim 5.0 simulator and DEAP framework with synthetic data traces and Nasa Ames/iPSC 860 real-world traces were used to evaluate the algorithm. From the evaluation results the proposed algorithm gives better results. We can observe 36.37% and 50.46% for makespan, 57.16%, 101.88% for system throughput, 1.47 and 0.94 differences of index improvement for degree of imbalance for synthetic and real-world traces respectively while having efficient cost values.