Label-affinity-Scheduler: Considering Business Requirements in Container Scheduling for Multi-Cloud and Multi-Tenant Environments

Abstract

The goal of this paper is to present a scheduling strategy that addresses a set of customer-specific business requirements when deploying containers, targeting multi-user and multi-cloud environments. This is done by extending the label scheme used to specify attributes for compute nodes and requirements for applications, and by associating workloads to the nodes with the highest affinity. The proposal is validated by implementing a custom scheduler for Kubernetes orchestrator. The custom scheduler was validated in an environment consisting of 25 nodes distributed across 4 providers with different hardware configurations and geographical locations. The results confirm the effectiveness of our scheduler in different scenarios, granting the business requirements assigned to each application.