Evaluation of container-scheduling algorithms for cluster management in the public cloud

Abstract

As the demand for efficient resource utilization in public cloud environments continues to rise, container scheduling algorithms play a pivotal role in optimizing the deployment and execution of applications. This study evaluates the efficacy of various container-scheduling algorithms, with a focus on determining the Stratus algorithm's ability to outperform Google's Borg algorithm on real-world cloud workloads run on Google Cloud. Through rigorous experimentation and data analysis, we find that the Stratus algorithm outperforms Google's Borg algorithm, especially in terms of the running cost of the instances - where we observed a 72% reduction in cost. However, the Stratus algorithm does not consider some intricacies of a cluster management setup, such as task constraints and dependencies. Hence, the Stratus algorithm is a more suitable algorithm for a customer looking to setup their own scheduling algorithm on the cloud, than for the purpose of cluster management for a cloud service provider.