Answers are grounded in this paper and cite their sources.
Background Existing simulators for container orchestration do not model the full Kubernetes stack—the scheduler, workload controllers, pod lifecycle, and autoscalers—together. Objective To provide CloudSim Plus‐K8s, a discrete‐event simulation extension of CloudSim Plus whose fidelity is set out in an explicit contract enforced by tests. Methods The extension models filter‐and‐score scheduling, the Horizontal and Vertical Pod Autoscalers and the Cluster Autoscaler, six workload controllers, and per‐service M/M/c latency. A 214‐test suite enforces the fidelity contract. We validated the simulator against kube‐scheduler‐simulator on 300 scenarios and against a four‐node Oracle Cloud k3s cluster running Google's Online Boutique. Results Filter and feasibility agreement with kube‐scheduler is 100%. Strict winner‐node agreement is 84% (5‐run mean; range 82%–87%), and combined winner‐or‐score‐set agreement is 100% on both substrates; steady‐state replica error is zero. The remaining strict disagreement arises solely from tie‐breaking among co‐optimal nodes. The M/M/c model reproduces traced p 95 latency within ±0.5 log 10 for one of eight operations under moderate load, with accuracy degrading under saturation. Conclusions CloudSim Plus‐K8s provides a falsifiable specification of the Kubernetes behaviours it reproduces, together with their tolerances, and is released as an open‐source, reproducible artefact.
No takes yet. Share an insight, caveat, or question.
Zarai et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: