Advancing OCP Telemetry Standards – Insights from a Drive Vendor

Kevin Brandt, Systems Architect at Micron Technology, details Micron's real-world implementation of OCP Data Center NVMe SSD telemetry standards, tracing the evolution from proprietary binary logs to the open-source nvme-cli OCP Plugin. He covers four major challenges Micron solved—defining human-readable debug, organizing data within the 32MB DA1-3 limit, minimizing latency of the C9h strings file pull, and educating customers—along with specific policies adopted (DA1 at 32KB, DA2 at 20-30MB, DA4 for maximal debug up to less than 1GB). Remaining challenges include no Windows nvme-cli support, out-of-band telemetry for drives not recognized as NVMe devices, and achieving human-readable core dumps in Data Areas 3 and 4.

Key Takeaways

Micron transitioned from proprietary NVMe SSD debug tooling to OCP open-source nvme-cli with the OCP plugin, achieving human-readable logs that customers willingly share and that align with internal failure signatures.
Micron's data-area policy maximizes DA1 to 32KB for fleet monitoring and DA2 to 20-30MB for failure analysis, keeps DA3 near zero, and uses DA4 up to less than 1GB for maximal debug—pushing all critical failure data into the decodable DA1/DA2 region.
A pre-pull methodology for the C9h Strings File, contributed by Micron to the OCP plugin, decouples it from time-critical telemetry collection and enables distributed debug where binaries are collected on one system and decoded on another.
Recommended nvme-cli for OCP plugin support: versions >= 2.10 required, 2.15 preferred; 2.11 must be avoided due to a segmentation fault; the plugin is absent on Windows blocking both log-pull and distributed decode workflows.
Three open proposals for the OCP community: a dedicated boot-negotiation telemetry profile for drives failing to enumerate as NVMe; PCIe Data Object Exchange as a faster out-of-band alternative to SMBus sideband (not yet a work item); and an industry-standard definition for human-readable core dumps in DA3/DA4.