Cost-Effective Scheduling of Massive Containers in Kubernetes

Explore cost-effective scheduling strategies for massive container deployments in Kubernetes through this conference talk by Yuan Chen from JD.com. Learn about JD's experience running one of the world's largest Kubernetes clusters in production, supporting diverse workloads from e-commerce to big data and machine learning. Discover how JD overcomes scalability and cost-effectiveness challenges through advanced scheduling techniques, including fine-grained resource usage monitoring, machine learning-driven allocation, mixed workload co-scheduling, and millisecond-level elastic scaling. Gain insights into Archimedes, JD's Kubernetes scheduling system, and its performance during JD's June 18 anniversary sale event, handling $24.7 billion in transactions. Delve into topics such as JDOS architecture, resource management challenges, bottleneck analysis, customization, preemption, admission control, group scheduling, and advanced scheduling techniques for CPU and memory optimization.

Intro
JD Retail Technological Infrastructure Group
JDOS (JD Data Center OS) Architecture
Resource Management: Challenges
A Typical JDOS Cluster
Bottleneck Analysis and Optimization
Customization and Optimization
Preemption and Admission Control
Group Scheduling and Binding
A Closed Loop System
Advanced Scheduling
Right Sizing of CPU Resource
Resource Usage and Performance Tradeoff
Memory Scheduling for Online Services
Memory Utilization of Online Services
Host Selection Optimization
Mixed Workload Placement
Hybrid Resource Pools
June 18 Anniversary Sale (2019)
Acknowledgements