Dynamic Pod Resource Boundary Adjustment in Web-Scale Clusters

Explore dynamic Pod resource limit adjustment strategies for web-scale clusters in this 38-minute conference talk from the Linux Foundation. Learn how to balance resource efficiency with application Service Level Objectives (SLOs) by co-locating Pods with different Quality of Service (QoS) classes on the same node and dynamically adjusting resource limits, especially during contention. Discover Alibaba Group's production cluster practices and lessons learned, which resulted in significant improvements: 14-30% increase in cluster resource usage, 76-87% reduction in tail latency (95th percentile), and 107-163% boost in transactions per second (TPS). Gain valuable insights on enhancing resource utilization and application performance in your own Kubernetes clusters using native approaches.

Dynamic Pod Resource Boundary Adjustment in Web Scale Clusters - Cheng Wang & Xiaoyu Zhang