Algorithms for scheduling scientific workflows on serverless architecture

Serverless computing is a novel cloud computing paradigm where the cloud provider manages the underlying infrastructure, while users are only required to upload the code of the application. Function as a Service (FaaS) is a serverless computing model where short-lived methods are executed in the cloud. One of the promising use cases for FaaS is running scientific workflow applications, which represent a scientific process composed of related tasks. Due to the distinctive features of FaaS, which include rapid resource provisioning, indirect infrastructure management, and fine-grained billing model a need arises to create dedicated scheduling methods to effectively use the novel infrastructures as an environment for workflow applications. In this paper we propose two novel scheduling algorithms SMOHEFT and SML, which are designed to create a schedule for executing scientific workflows on serverless infrastructures concerning time and cost constraints. We evaluated proposed algorithms by performing experiments, where we planned the execution of three applications: Ellipsoids, Vina and Montage. SDBWS and SDBCS algorithms were used as a baseline. SML achieved the best results when executing Ellipsoids workflow, with a success rate above 80%, while other algorithms were below 60%. In the case of Vina, all the algorithms, except SDBWS, had a success rate above 87.5% and in the case of Montage, the success rate of all algorithms was similar, over 87.5%. The proposed algorithms' success rate is comparable or better than offered by other studied solutions.