Explore strategies for optimizing resource utilization in Kubernetes clusters by co-locating CPU and GPU workloads. Learn how Ant Financial and Alibaba achieved a 10% increase in utilization through innovative approaches. Discover the creation of a new QoS class, implementation of node-level cgroups for batch jobs, and use of PodGroup CRD for gang scheduling. Gain insights into building and managing a co-location cluster with over 100 GPU and 500 CPU nodes, effectively combining long-running services and AI batch jobs. This 37-minute conference talk from the Linux Foundation provides valuable experience and practices for maximizing resource efficiency in Kubernetes environments.
Co-Location of CPU and GPU Workloads for High Resource Efficiency in Kubernetes
Overview
Syllabus
Co-Location of CPU and GPU Workloads with High Resource Efficiency - Penghao Cen & Jian He
Taught by
Linux Foundation
Tags
Related Courses
-
Coordinate Workloads Colocation: QoS-Oriented Scheduling Enhancement on Kubernetes
-
Improving GPU Utilization and Accelerating Model Training with Kubernetes Scheduling Framework and NRI
-
Precision Matters: Scheduling GPU Workloads on Kubernetes
-
Minimizing GPU Cost for Deep Learning on Kubernetes
-
Maximizing GPU Utilization Over Multi-Cluster - Challenges and Solutions for Cloud-Native AI Platform
-
Efficient Multi-Cluster GPU Workload Management with Karmada and Volcano
