Cost Optimization and Resource Management in an Online Education Platform: From XEN Migration to Container‑Based Scaling

The background explains the high "burn‑money" problem in the internet industry, where not only marketing spend but also the cost of machines, cloud services, bandwidth, and IDC equipment constitute a major expense for large‑scale online education services.

To avoid single‑point failures, each service requires N+1 resources, leading to over‑provisioning during peak periods and low‑utilization machines after traffic subsides. The article outlines the challenges of expanding capacity, ensuring identical machine and software configurations, and the risks associated with XEN instability that prompted a migration to KVM.

Two internal platforms were developed: the "Hummingbird Platform" for visualizing resource trees and the "Online School Cloud Platform" built on Kubernetes to increase machine utilization. The cloud platform enables containerized deployments, providing consistent runtime environments via Docker images and allowing seconds‑level container start‑up compared to tens of seconds for VMs.

Dynamic scaling is achieved through manual replica adjustments and Kubernetes HPA‑based automatic scaling, giving developers the ability to set desired replica counts and let the system handle rapid scaling up or down. Health‑checking components such as HChecker replace gateway probes to ensure failed pods are removed from service endpoints.

Resource auditing and cost‑optimization steps include precise asset classification, reclaiming unused servers, standardizing OS and software versions, and establishing a closed‑loop review process. The initiative resulted in the decommissioning of nearly a thousand XEN VMs, migration of core services to KVM, and a reported savings of about ten million RMB in the first half of 2020.