Next-Gen AI Orchestration With Volcano On Kubernetes
summary
Volcano positions itself as a unified AI scheduler that moves beyond batch to coordinate multiple schedulers and workload types, implementing queue and gang scheduling and a new agent scheduler to handle bursty, latency‑sensitive agent tasks concurrently with batch training. It models network domains with a HyperNode CRD (auto‑discovery for NVIDIA FAB and node labels) to surface switch bandwidth and congestion for topology‑aware placement. It integrates LLM serving stacks like Kthena and vLLM, and ships Casino (token‑aware autoscaling, KB cache, PD‑group routing) and AgentCube (agent primitives + SDK) while adding GPU sharing, DRA, and heterogeneous pooling.
AI-written summary · about
related talks
30mVolcano: A Unified Scheduling Platform for Cloud Native AI
43mRun Your AI Workloads and Microservices on Kubernetes More Easily and Efficiently...
30mCloud Native AI: Harness the Power of Advanced Scheduling for High-Performance AI/ML Training
28mEfficient Scheduling Of High Performance Batch Computing For Analytics Workloads With Volcano
26mIntro: Kubernetes SIG
32mVolcano: Intro & Deep Dive
24mIntro To Volcano: Cloud Native Batch System
37mImproving Performance of Deep Learning Workloads With Volcano