Humboldt-Universität zu Berlin - Mathematisch-Naturwissenschaftliche Fakultät - Institut für Informatik

Verteidigung Bachelorarbeit: Lennart Seiffer

Bachelor thesis on How Workflow Management Systems address Kubernetes - A Taxonomy

Dear Members of the Institute

you are cordially invited to the defense of Lennart Seiffer regarding his Bachelor thesis on

How Workflow Management Systems address Kubernetes - A Taxonomy

The defense will take place Monday, 29.8.2022, 10.00 s.t.,  via Zoom :

https://hu-berlin.zoom.us/j/67169152360?pwd=cjh1a1FFeGxPY1F1T29lV2xoTm9BZz09

Best wishes,
Ulf Leser

------------------

Abstract
Data analysis workflows are an integral part of modern research in the natural sciences. Yet, several issues are slowing down the process of creating and sharing workflows. One issue is the lack of portability and reproducibility of workflows across different environments. Several workflow management systems aim to achieve these properties by abstracting their execution over various infrastructures. One common attempt to enable this abstraction is the use of resource managers like Kubernetes. Kubernetes manages compute nodes and storage in a uniform
way, simplifying the migration process of workflows across different Clouds and datacenters.
In this thesis, we give an overview of the utilization of Kubernetes by multiple state-of-the-art workflow management systems. We aim to categorize the systems, based on the way they address Kubernetes. To discuss aspects of workflow implementation, performance, and distribution of responsibilities, we adapt a simplified word count workflow to all systems. For evaluation, we run each workflow in a Kubernetes cluster hosted in the Cloud, measure the execution time, and track logs of Kubernetes events and the workflow executors. We use the logs, along with documentation and further metadata, to investigate how
the executors are implemented. Based on our findings, we discuss dvantages and disadvantages of the different approaches towards using Kubernetes as a resource manager.