USE Method

Overview

The USE method, from Brendan Gregg, is a diagnostic checklist for resources. For each resource, ask three questions: how busy is it (Utilization), is work waiting (Saturation), and is it failing (Errors). It turns a vague performance hunt into a bounded sweep.

How it works

Enumerate resources, then check each metric type:

Resource      Utilization            Saturation              Errors
CPU           cpu busy per core      run queue length        machine check events
Memory        used vs total          swapping, OOM kills     allocation failures
Disk          device busy percent    I/O queue depth         device errors
Network       bytes vs line rate     drops, retransmits      interface errors
DB pool       in-use / max           callers waiting         timeouts
Threads       busy / total           task queue depth        rejected tasks

Interpretation rules: utilization near 100 percent usually means a bottleneck; any sustained saturation means queued work and added latency; any errors deserve investigation. Because the sweep is finite, you can finish it, and you will have covered resources nobody thought of.

Pitfalls

• Reading averaged CPU across cores. A single-threaded workload can saturate one core at 12 percent average.
• Treating 70 percent disk utilization as safe. For a device that serves requests serially, queueing begins well below 100 percent.
• Forgetting virtual limits such as cgroup CPU throttling, file descriptors, and ephemeral ports.
• Ignoring errors because they look rare. A small count of retransmits can explain tail latency.

Key takeaways

• Check utilization, saturation, and errors for every resource.
• Saturation is the early warning that utilization misses.
• Include software resources such as pools and locks.
• Use USE for resources and RED for services.