Volcano: Orchestrating the Full AI Lifecycle – From Training To Inferen... Chen Zicong & Hajnal Máté
About this talk
This talk explores how Volcano, developed by Huawei Technologies and Aumovio, serves as a Unified Scheduling Platform capable of orchestrating the full AI lifecycle, from training to inference and agent workloads. The speakers highlight how Volcano addresses infrastructure fragmentation by delivering high throughput scheduling and enabling multi-scheduler coordination for diverse workloads beyond batch jobs. Key topics include the capabilities of Volcano-Global for splitting massive training jobs across clusters, Kthena for enterprise-grade LLM serving, and AgentCube for rapid agent workload scheduling. Additionally, the session covers the modern resource abstraction offered by Volcano through DRA integration, GPU sharing, and heterogeneous pooling to improve task-to-accelerator mapping efficiency.
More from this event
See all 436 talks →
Best of KubeCon + CloudNativeCon Amsterdam 2026
2:17
The Quiet Work of Forever: Sustaining Open Source Communities - O. Hope Amaechi-Okorie, JSON Schema
26:24
Evolving KServe: The Unified Model Inference Platform for Both Predictive and... F. Spolti & J. Lee
32:40
Preventing S3 Cost Storms: Applying Cortex’s Efficiency Lessons to I/O-Heav... A. Fishman-Lichterman
5:32