Diagnose Storage I/O Performance and Identify the Workload

When a service is slow and storage I/O is suspected, explain how you would use system metrics (such as iostat), identify the offending workload, and confirm whether the storage path is the bottleneck.

Constraints & Assumptions

No device type or filesystem is specified. Queueing, latency, and utilization must be interpreted in the context of the actual storage stack, not by assuming a universal threshold for every disk or SSD.

Clarifying Questions

Are reads, writes, or both slow? Is the access pattern random or sequential? What are the request sizes, concurrency, caching layers, and durability requirements? Does the application block on sync operations or metadata updates?

What a Strong Answer Covers

Device-level latency and throughput, process attribution, cache effects, workload shape, and a data-driven remediation plan.

Follow-up Questions

Why is a high utilization percentage insufficient to prove saturation on every device? How would you separate application-level queueing from device service time? What changes when storage is remote?

Overview: Diagnose storage slowness using device metrics, process attribution, workload characteristics, cache and durability effects, and measurements suited to the storage stack.