Myberk Ozerk

Overview

Myberk Ozerk is a storage reliability engineer at Microsoft who represents Microsoft's operational requirements in the OCP data center storage specification effort. At the October 2025 OCP Global Summit he presented Microsoft's reliability-first perspective and the end-to-end self-healing pipeline the company has built around OCP telemetry capabilities.

Reliability is Microsoft's top priority for SSD fleet management. Ozerk quantified the failure landscape: firmware asserts and hard hangs together account for roughly 90% of failures across Microsoft's storage fleet. The OCP spec's full-system-resiliency features directly address these dominant failure modes. Microsoft's approach to reliability combines real-time online telemetry and post-failure offline log extraction. Online telemetry is collected from NVMe data area one on an hourly and daily cadence; after a failure event, offline log extraction supplements this data. The combined stream feeds detection, prediction, and root-cause analysis workflows that can ultimately trigger firmware deployment decisions.

The firmware lifecycle is a critical constraint on how quickly identified issues can be resolved. Ozerk cited a 6-to-9-month timeline covering development, validation, qualification, and deployment of any given firmware feature. During that window, Microsoft relies on its prediction capabilities to mitigate known failure patterns — particularly ground defects and end-of-life drive behavior — before the firmware fix is available.

Microsoft's detection and mitigation pipeline parses drive, server, and kernel logs into a central database, then applies both expert-rule static thresholds and ML models. The mitigation layer supports a range of responses proportional to severity: live migration, customer notification to vacate nodes, and full service healing where virtual machines must be rebooted.