推进OCP遥测标准——来自驱动器供应商的见解.pdf

1、Kevin BrandtAdvancing OCP Telemetry Standards Insights from a Drive VendorAdvancing OCP Telemetry Standards Insights from a Drive VendorKevin BrandtStorageBeginningsMicron started our telemetry journey with NVMe products using proprietary debug logs,proprietary tools to pull the logs and reliant on

2、binary data from customers.CurrentRelying on OCP defined telemetry logsOpen-source tools:nvme-cli/OCP Plugin Micron contributed changes to the OCP pluginHuman Readable output from customersTelemetry JourneyWhat did we learn on this journeyOCP MarketplaceOCP Data Center NVMe SSD-Telemetry JourneyClou

3、d NVMe 1.0DC NVMe SSD 2.0DC NVMe SSD 2.5DC NVMe SSD 2.6SEC-22“Human Readable Debug”SEC-19 VS logs disabledTelemetry logs DA1-10ms latency impactBackground telemetry cant impact latencyDA1/DA2 requiredDA1 Size 16KDA1 Latency Impact 1ms typ,5ms maxStatistic identifiers addedNVMe 2.0(DA4 optional)“Huma

4、n Readable Debug”via OCP PluginC9h Strings FileTelemetry profilesTEL-11 Telemetry sufficient for debugSEC-22 DA1/DA2 Human Readable and sufficient for debugDA1 Size 32KDA1 Size VSDA1 Latency impact 1ms typ,10ms maxDA2/3 optionalTelemetry AEN clarifications2020202120232024What is“human readable”debug

5、 and how to enable human readable debug?Data limits and organizationLatency impacts of pulling telemetry logsEducating customers on how to use the debugMajor Challenges in Meeting OCP TelemetryWhat is“human readable”debug?As the OCP spec evolved and SSD vendors asked questions this became more clear

6、As OCP 2.5 was being developed Micron decided we would proceed with planning to support decode via OCP CLIMajor Challenge:Human ReadableDatacenter NVMe SSD Specification 2.5-2023Datacenter NVMe SSD Specification 2.6-2024NVMe Cloud SSD Specification 1.0-2020As OCP evolved a key question was“How shoul