Evaluation of Function as a Service for Cloud Workflows

Scientific Workflows assist with automating tasks in many areas. Orchestrating the execution of these workflows requires the use of Workflow Management Systems. This can be realized through different cloud infrastructures. One such infrastructure model is Function as a Service (FaaS), a concept where the user does not have to manage servers and resources and can focus on writing the code itself. This thesis aims to investigate the use of FaaS for executing cloud workflows and provide a quantitative and qualitative evaluation. For this purpose, we describe the process of deciding on a FaaS framework fitting our use case by comparing several open-source frameworks. We also share our experience implementing a FaaS approach with the Workflow Management System Steep, using the OpenFaaS framework. Additionally, we implement an alternative method of execution with Kubernetes. We perform experiments with a real-world workflow to compare the execution times of our two newly added solutions with Steep’s currently available Docker execution. Also, we research the scale-to-zero feature provided by OpenFaaS. The findings show both of our solutions work with reasonable execution time. We discuss the advantages and limitations of FaaS based on our experiment results and implementa- tion experience, concluding that the configuration and deployment overhead makes FaaS more suited for experienced developers. On the other hand, Kubernetes is a promising alternative for users looking for a more straightforward option with less setup overhead. To our knowledge, this study is the first to do both a quantitative and a qualitative evaluation of FaaS. We do not provide an optimized serverless solution but rather present a basis for further research in the field. Future work will focus on improving our solution and eventually automating the choice of the execution setup.