Characterizing AFCL Serverless Scientific Workflows in Federated FaaS

This paper introduces several, publicly available, serverless scientific workflows Montage, BWA, and Monte Carlo developed at a high level of abstraction using the Abstract Function Choreography Language (AFCL). Any individual function can run across federated FaaS comprising cloud regions of AWS and GCP. We present the support for composition with AFCL and execution with the xAFCL serverless workflow management system. For each AFCL workflow, we present implementation details, networking, and complexity. The evaluation of the presented serverless workflows shows that workflow functions download ephemeral data and run computation faster on AWS than on GCP. However, functions on GCP upload faster on the collocated storage.