QoS-Aware Multi-Tenant Container Orchestration Using Kubernetes

Abstract

In modern cloud-native environments, container orchestration platforms such as Kubernetes play a pivotal role in managing microservices-based applications at scale. However, the advent of multi-tenancy—where multiple independent applications share the same physical cluster—introduces complex challenges in guaranteeing Quality of Service (QoS). Resource contention, noisy-neighbor interference, and unpredictable workload patterns can lead to performance degradation, SLA violations, and tenant dissatisfaction. This manuscript presents an enhanced QoSaware multi-tenant orchestration framework built on Kubernetes primitives, augmented by custom controllers and a scheduler plugin. We define a formal model for tenant Service Level Objectives (SLOs) via a SloPolicy Custom Resource Definition, encompassing latency percentiles, throughput requirements, and weighted resource shares.  A QoS-SLO Controller reconciles policies into Kubernetes constructs—PriorityClasses, ResourceQuotas, and LimitRanges—while a Scheduler Plugin computes real-time QoS scores to guide scheduling and preemption decisions. A closed-loop feedback mechanism, leveraging Prometheus metrics, dynamically adjusts priorities and autoscaling parameters upon SLO deviations. Experimental evaluation on a heterogeneous microservices workload demonstrates that our framework reduces 99th-percentile latency by up to 45%, cuts SLA violation rates to under 5%, and improves fairness in resource allocation without sacrificing overall cluster utilization (maintained above 70%). These results underscore the viability of integrating declarative SLOs with adaptive scheduling for robust multi-tenant orchestration.