Squeeze Your K8s - Adopting Time-Series Forecasting Models in FinOps Practices

Explore how Shopee adopts time-series forecasting models in FinOps practices to optimize Kubernetes resource utilization in this 38-minute conference talk. Learn about the challenges of static CPU/Memory specs in Kubernetes and how they lead to over-provisioning and low cluster utilization. Discover the potential of dynamic resource specs based on actual usage, made possible by Kubernetes' support for in-place resource resizing since version 1.27. Delve into Shopee's approach to predicting POD resource utilization accurately using different time series forecasting models for long-term and short-term predictions in scheduling and descheduling scenarios. Understand the advantages and suitable scenarios for various models, as well as the process of training and tuning these models for Kubernetes cases. Gain insights into leveraging inference results for service personas to significantly overcommit resources without affecting service latency. Uncover how this project contributes to substantial server cost savings and gather inspiration for your own FinOps practices in cloud native environments.