openEuler Wants Linux to Treat an Entire AI Cluster Like One Machine

At Open Source Summit Europe 2026 in Prague, openEuler pitched SuperPoD OS: a Linux kernel that manages compute, memory and accelerators across many nodes as a single resource pool. It claims node-failure notification within 500 ms, 20% lower RPC and RH2D latency, and agent sandbox cold starts of 20 ms.

• SuperPoD OS pools CPUs, NPUs, GPUs and memory across many nodes
• Node-failure notification latency is 500 ms or less, with SEA fault-storm suppression
• UMDK and an async pipeline cut RPC and RH2D latency by 20%
• Agent sandboxes: 20 ms cold starts and over 2,000 concurrent environments