1. Background Research

REANA is as a free open-source platform designed to execute reproducible declarative data analyses on containerized compute clouds. With Kubernetes as its primary backend, it efficiently manages computational workflows through the inherent Kubernetes Job API, stream- lining user job scheduling. The central goal of this openlab summer student project is directed into assessing the performance of contemporary Kubernetes batch scheduling systems like Armada, Kueue, and Volcano. We concentrated our efforts on Kueue due to its status as the Kubernetes-native scheduling solution being actively developed by the Kubernetes community. Through the use of an architecture that emulated REANA workflow submissions, we have gauged and compared scheduling solutions, focusing on performance benchmarks and supplementary capabilities like equitable resource allocation, dynamic resource adaptability, and scalability capabilities under many thousands of workloads. This evaluation aimed to assess the suitability of Kueue as a possible batch scheduler in the REANA reusable analyses platform.

!s41567-018-0342-2.pdf

!fdata-04-661501.pdf!epjconf_chep2018_06034.pdf!chep2023_atlas_pmssm_reana_paper.pdf!Evaluating_Kubernetes_batch_scheduling_systems_for_containerized_declarative_data_analyses.pdf