A Comparative Analysis of Kueue, Volcano, and YuniKorn
Talkat KubeCon + CloudNativeCon Europe 2025 · on Wednesday, 2025-04-02· at 11:15· Level 1 | Hall Entrance N10 | Room G·
summary
Compares Kueue, Volcano, and Apache YuniKorn with concrete benchmarks: 10,000-pod workloads across patterns (many single-pod jobs, few multi-pod jobs, gang scheduling on/off) on Kubernetes 1.32, collecting events from API server audit logs. Finds Volcano wins gang scheduling latency and job completion; Kueue often wins non-gang high-job-count cases thanks to its suspension pattern and lighter admission overhead; YuniKorn uses placeholder pods (doubling API ops) and limits preemption to inter-queue scope. Benchmarks and tooling are open source.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:06Why multiple schedulers exist
- 3:18History of Kueue, Volcano, YuniKorn
- 5:03Kubernetes integration
- 7:57Multi-queue resource management
- 11:34Gang scheduling differences
- 14:28Framework integration
- 17:03Cluster autoscaling
- 19:07Benchmarking methodology
- 22:47Benchmark results without gang
- 26:28Benchmark results with gang
- 29:53Conclusion and roadmap
- 30:38Q&A
related talks
30mApache YuniKorn A Kubernetes Scheduler Plugin for Batch Workloads
41mApache Yunikorn: State of Union and Future
26mIntro: Kubernetes SIG
32mVolcano: Intro & Deep Dive
36mNext Level Spark on Kubernetes with Apache YuniKorn (Incubating)
29mScale Your Batch / Big Data / AI Workloads Beyond the K8s...
24mVolcano – Cloud Native Batch System for AI, BigData and HPC
24mIntro To Volcano: Cloud Native Batch System