Coordinate Workloads Colocation: QoS-Oriented Scheduling Enhancement on Kubernetes

Explore a conference talk on enhancing QoS-oriented scheduling in Kubernetes for efficient workload colocation. Dive into the limitations of current Kubernetes QoS classes and learn about Koordinator, an innovative solution that addresses these challenges. Discover how Koordinator implements sub-QoS classes, utilizes dynamic metrics for improved resource overcommitment, and applies fine-grained resource orchestration to solve noisy neighbor problems. Gain insights into the architecture, key design features, and various scheduling enhancements that optimize resource efficiency and performance for complex workloads in co-location scenarios. Understand the benefits of Koordinator for Spark on Kubernetes, resource overcommitment, load-aware scheduling, and CPU management. Learn about the project's milestones and how to get involved in this cutting-edge development for cloud-native environments.

Intro
Contents & Speaker
Resource Efficiency
Resource Model in K8s
Why Noisy Neighbors
Key Features
Architecture
Key Design - Koordinator Priority
Key Design - Koordinator QoS
QoS and Priority
Spark on Kubernetes
Spark on K8s with Koordinator
Resource Overcommitment
Load-aware Scheduling
Fine-grained CPU Management
koordlet CPU Suppress
koordlet Satisfaction Eviction
Various QoS Management
Various Scheduling Features
Koordinator Milestone
Getting Involved