OCP HM - OpenRMC-DM Project Call (Dec 09, 2025)

Why It Matters

Standardizing bulk management and domain‑level abstractions enables OCP’s AI‑focused data centers to scale efficiently, reducing operational complexity and ensuring consistent firmware and telemetry across multi‑rack deployments.

Key Takeaways

• Added single‑command inventory retrieval for all systems in a rack.
• Introduced group operations: BIOS, boot order, telemetry configuration.
• Debated logical vs physical rack definitions and proposed “scale‑up domain.”
• Pod manager layer added to aggregate multiple racks and provide failover.
• Consensus to expand telemetry and standardize domain‑centric interfaces.

Summary

The final 2025 OpenRMC‑DM call centered on the upcoming 1.3 specification, which expands Redfish‑based management to support bulk inventory, firmware versioning, and group operations across an entire rack. Participants reviewed new Git changes that introduce single‑command queries for CPU, power, and firmware data, as well as bulk configuration of BIOS, boot order, telemetry, and event subscriptions. Key insights included the need for filters to target specific firmware components, the challenge of defining a rack as either a physical enclosure or a logical collection, and the proposal to introduce a “scale‑up domain” identifier that abstracts multiple racks under a single management view. The discussion also covered the addition of a second‑layer pod manager to list racks, aggregate power consumption, and provide failover capabilities. Notable remarks highlighted the tension between existing terminology and emerging architectures: “Physical rack versus logical rack can be confusing; a domain ID may clarify multi‑rack deployments,” and “Telemetry must be richer to make pods and domains intelligent.” The team agreed to flesh out telemetry use cases and to rename or augment concepts such as pod manager and domain for broader industry adoption. The implications are significant for OCP members building AI‑infer clusters: standardized bulk commands and domain‑centric interfaces will simplify large‑scale provisioning, firmware consistency, and health monitoring across heterogeneous hardware, accelerating deployment of high‑density compute pods.