Optimizing Kubernetes Scheduling : Balancing Cost, Performance, and Resiliency

Sammanfattning

With the development of cloud software, industry is increasingly migrating their applications from on-premise infrastructure to cloud environments. Kubernetes is a widely used container orchestration platform, which has become the standard for deploying and managing containerized applications. One of the core responsibilities of Kubernetes is job scheduling, which involves allocating physical resources to containers efficiently. Despite its maturity and advanced capabilities, Kubernetes primarily relies on a greedy scheduling approach that considers only CPU utilization, often leading to suboptimal resource allocation. This limitation can result in unnecessary spending on underutilized compute resources, imbalanced workloads, and reduced system resiliency. Addressing these challenges is crucial for organizations aiming to optimize performance and cost efficiency in cloud-native deployments. In this work, we propose a heuristic Ant Colony Optimization (ACO)-based scheduling algorithm that jointly optimizes for cost efficiency, performance, and resiliency in Kubernetes environments. Our algorithm is evaluated using the kube-scheduler-simulator and compared against two benchmarks: the default Kubernetes scheduler and the industry-standard Pod Anti-Affinity strategy. Experimental results show that our ACO-based scheduler consistently outperforms the baselines across key metrics, producing more balanced placements in terms of resource cost, load balancing, and resiliency.