Hua Liu
Tencent Cloud Native Technologist
Tencent expert engineer, responsible for the technical exploration and landing work of cloud native technology in the field of big data and AI, currently focusing on the use of the underlying runtime technology to improve the stability and reduce the cost of big data and AI business.
Topic
Peak Thermal Migration: CPU/GPU Load Thermal Migration Practices for K8s Production Environments
Introduction: Container thermal migration has a promising application in the K8S ecosystem. When combined with low-priced unstable resources, container hot migration technology allows consumers to obtain similar compute power at a lower price, thereby reducing costs significantly. Container hot migration also allows for greater flexibility in lossless rescheduling to better balance cluster resources and reduce hotspots and resource fragmentation. Outline: The Kubernetes community's discussion of container hot migration dates back to before the release of Kubernetes 1.0, and is a huge topic involving the operating system, runtime, container network, and Kubernetes APIs, which has seen little progress in the Kubernetes community to date. In this talk, we will provide an introduction to Tencent's internal system that performs an average of 20,000 container hot migrations per day around the above issues, and discuss our ongoing GPU hot migration.
