A Multi-Objective Optimisation Framework for Balancing Performance, Cost and Reliability in Containerised Applications

Abstract

Modern cloud-native platforms rely heavily on containerized microservice architectures to achieve continuous delivery, rapid scalability, and operational modularity. However, managing containerized applications across heterogeneous cloud environments imposes conflicting operational pressures: minimizing service-level agreement response latencies, curtailing cloud infrastructure expenditure, and maximizing system reliability under unpredictable traffic bursts. Existing autoscalers, such as the standard Kubernetes Horizontal Pod Autoscaler, rely upon reactive threshold heuristics that lead to over-provisioning during steady states and catastrophic latency degradation during sharp demand surges. This paper proposes a comprehensive multi-objective optimization framework that balances end-to-end performance, infrastructure cost, and service reliability simultaneously. Our approach integrates an evolutionary many-objective Pareto optimization engine with predictive cloud telemetry forecasting. By evaluating non-dominated Pareto frontiers across dynamic resource configurations, the framework adaptively schedules pod replicas, modulates cpu and memory quotas, and routes traffic across spot and on-demand instances. We evaluate our framework on a multi-tier microservice testbed under real-world diurnal request traces. The empirical results prove that our framework reduces 99th percentile response latency by 34.2%, curtails cloud operating expenditures by 28.6%, and eliminates service-level agreement violations under aggressive burst loads.