SLO-Oriented Scheduling With Resource Estimation to Optimize Utilization

SLO-Oriented Scheduling With Resource Estimation to Optimize Utilization

Explore a conference talk on SLO-oriented scheduling with resource estimation for optimizing utilization in Kubernetes clusters. Learn how Alibaba manages hybrid workloads to achieve high resource utilization while meeting various Service Level Objectives (SLOs). Discover key enhancements including a real-time metrics pipeline, an efficient algorithm for estimating resource consumption patterns, and an SLO-oriented scheduling policy for opportunistic pod scheduling. Gain insights into practical implementations and performance statistics from production clusters handling diverse workloads such as e-commerce services and business-critical AI jobs.