// the find
brendangregg/pmc-cloud-tools
PMC (Performance Monitoring Counter) tools for the cloud
Linux shell wrappers around hardware performance counters that report IPC, branch misprediction, cache hit ratios, and TLB walk cost per interval. They are aimed at engineers investigating production CPU behaviour on EC2 instances that expose PMCs, who want those numbers without writing perf invocations by hand.
pmcarch answers the first questions on a slow box in one command: cycles, instructions, IPC, branch mispredict rate, and LLC hit ratio, sampled at a chosen interval and scoped with -C, -p, or -c. The README splits the architectural counters (pmcarch) from the Nitro-only extended set (tlbstat) and names the instance families that expose each, so you don't run a tool on hardware that can't feed it. The derived columns are defined (K_CYCLES as x1000, BMR% and LLC% as ratios), so the output can be read without opening the source.
The last push was December 2021, and the README calls these works-in-progress that may not work on the cloud yet, so nothing here has been checked against recent kernels or perf releases. Everything depends on named PMC event aliases and Intel-family counters that vary by CPU generation and hypervisor, and there is no tested-instance list or fallback when an event is missing. The README has no install or dependency section and documents only the fixed-width table output, with no machine-readable mode for piping into other tools.