Scale Kubernetes to Manage 60K Nodes Through Architectural Extension

Overview

Explore how to scale Kubernetes to manage 60,000 nodes through architectural extension in this 50-minute conference talk by Ying Huang and Ying Xiong from Futurewei Technologies. Dive into the Arktos project, its scale-out architecture, multi-tenancy design, and unified runtime infrastructure. Learn about critical performance improvements, optimization techniques, and resource utilization in large-scale Kubernetes clusters. Gain insights into the Centaurus project and its relationship with Arktos. Analyze performance test results and understand the tools used for evaluation. Discover how to enhance event watch optimization and improve master resource utilization in both TP1 and RP1 Arktos clusters.

Syllabus

Intro
Outlines
WHO ARE WE AND WHAT IS ARKTOS PROJECT
WHY CENTAURUS AND ARKTOS
ARKTOS INTRODUCTION
ARKTOS SCALE-OUT ARCHITETCURE
ARKTOS MULTI-TENANCY DESIGN
ARKTOS UNIFIED RUNTIME INFRASTRUCTURE
Arktos Scale-Out Architecture
Arktos Optimization
Performance Test Results & Analysis
Performance Test Tool
Critical Changes for Performance Improvement
Events Watch Optimization
Final Performance Data
Arktos Cluster TP1 Master Resource Utilization
Arktos Cluster RP1 Master Resource Utilization