SLO-Oriented Scheduling With Resource Estimation to Optimize Utilization - Yu Yinghao & Wang Shuang

Don’t miss out! Join us at our upcoming events: EnvoyCon Virtual on October 15 and KubeCon + CloudNativeCon North America 2020 Virtual from November 17-20. Learn more at https://kubecon.io The conferences feature presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.

SLO-Oriented Scheduling With Resource Estimation to Optimize Utilization - Yu Yinghao & Wang Shuang, Alibaba

A great variety of workloads are surging into the Cloud. As the de facto operating system of the Cloud, the Kubernetes eco-system is soon faced with this challenge: How to manage hybrid workloads to achieve high resource utilization while accomplishing variant SLOs？ In this talk, we will share practices of consolidating hybrid workloads, including e-commerce services and business-critical AI jobs, in Kubernetes clusters of Alibaba. We elaborate on critical details of the enhancements we have made: 1. a real-time pipeline to stream the cluster metrics into the the scheduling system; 2. an efficient algorithm estimation of the upcoming resource consumption patterns; 3. an SLO-oriented scheduling policy that opportunistically schedules best-effort pods leveraging allocated but not used resources. Lastly, we'll share statistics of how the entire system works in production clusters.

https://sched.co/cp9L