Scientific workflow scheduling algorithms in cloud environments: a comprehensive taxonomy, survey, and future directions

Scientific workflows are large applications that consist of smaller computational units called tasks that have data dependency on each other. The tasks of a workflow can be scheduled and executed on distributed resources in a parallel manner. Cloud computing offers distributed, scalable, virtualized, cost-effective computing environments making them ideal platforms to execute scientific workflows. Cloud services provide their users with a vision of an unlimited amount of computing resources. However, considering different types of resources and QoS requirements, the problem of workflow scheduling lies in the NP-complete class. Thus, numerous types of research have been conducted in this area during the past years. In this paper, we aim to provide a comprehensive study of the workflow scheduling problem, existing solutions, and available tools that can be used by researchers in this domain. First, we present a taxonomy on scheduling algorithms and examine the existing works from different perspectives from application and resource models to algorithms’ objectives and their nature. We also have presented a taxonomy of evaluation data sets as well as simulation tools and their architecture since the evaluation of an algorithm is important and must be performed accurately. Next, we survey some of the most recent works in the context of the proposed taxonomy with a focus on emerging cloud services like serverless computing or workflow as a service platform and state-of-the-art scheduling approaches. Moreover, we discuss some of the existing gaps in the literature and identify possible research directions that can be seen as potential contributions to future developments.