A Heuristics-Based Cost Model for Scientific Workflow Scheduling in Cloud

Scientific Workflow Applications (SWFAs) can deliver collaborative tools useful to researchers in executing large and complex scientific processes. Particularly, Scientific Workflow Scheduling (SWFS) accelerates the computational procedures between the available computational resources and the dependent workflow jobs based on the researchers’ requirements. However, cost optimization is one of the SWFS challenges in handling massive and complicated tasks and requires determining an approximate (near-optimal) solution within polynomial computational time. Motivated by this, current work proposes a novel SWFS cost optimization model effective in solving this challenge. The proposed model contains three main stages: (i) scientific workflow application, (ii) targeted computational environment, and (iii) cost optimization criteria. The model has been used to optimize completion time (makespan) and overall computational cost of SWFS in cloud computing for all considered scenarios in this research context. This will ultimately reduce the cost for service consumers. At the same time, reducing the cost has a positive impact on the profitability of service providers towards utilizing all computational resources to achieve a competitive advantage over other cloud service providers. To evaluate the effectiveness of this proposed model, an empirical comparison was conducted by employing three core types of heuristic approaches, including Single-based (i.e., Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Invasive Weed Optimization (IWO)), Hybrid-based (i.e., Hybrid-based Heuristics Algorithms (HIWO)), and Hyper-based (i.e., Dynamic Hyper-Heuristic Algorithm (DHHA)). Additionally, a simulation-based implementation was used for SIPHT SWFA by considering three different sizes of datasets. The proposed model provides an efficient platform to optimally schedule workflow tasks by handling data-intensiveness and computational-intensiveness of SWFAs. The results reveal that the proposed cost optimization model attained an optimal Job completion time (makespan) and total computational cost for small and large sizes of the considered dataset. In contrast, hybrid and hyper-based approaches consistently achieved better results for the medium-sized dataset.