Running Mixed Workloads on Kubernetes at the Institute for Health Metrics and Evaluation
summary
IHME integrates traditional HPC (15,000 cores, petabytes of inputs, 30 TB final outputs) and containerized services on a single Kubernetes cluster using Univa Navops, replacing the default scheduler with a Grid Engine path to enable mixed HPC+containers and virtual multi‑tenancy. The pilot runs mostly sub‑100‑process jobs, adds preemption and dynamic priorities, simplifies administration, and lifts cluster utilization from <20% to >50%, so the combined environment is working and still under active tuning.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 4:00Data Source
- 7:00Goldberg Disease
- 8:08Current Work
- 9:43Burden of Disease
- 10:50Visualization Tools
- 11:23Policy Impact
- 13:00Partner Organizations
- 14:54The Team
- 18:38Barebones
- 19:54HPC Cluster
- 20:18Software
- 22:42Architecture
- 23:06Moving to Kubernetes
- 24:56Virtual Multitenancy
- 28:19Questions
related talks
23mImplementation Challenges: From HPC to Containers in the Academy
31mManaged Kubernetes — Next Gen Academic Infrastructure?
19mKeynote: Building Bridges: Cloud Native and High Performance Computing
13mOptimizing Workload Management in a Multi-Cluster Environment: A Case Study
20mA full-scenario colocation of workloads based on Kubernetes
41mBuilding & Operating a Large-scale HPC AI Cluster on Kubernetes
41mContainer engines in High Performance Computing: some usage stories from CINECA
20mBenchmarking HPC workloads on OpenShift