Skip to main content

Clock Frequency Tiering Validation

These findings came from the optional clock-lane experiment after comparing a stock run to a clock-tiered run.

warning

This is the dedicated-lane experiment, not the shared-pool contention path.

What Was Tested​

  • kvm-worker-01 on CPUs 2-4 with a 3 GHz ceiling
  • kvm-worker-02 on CPUs 5-7 with a 2 GHz ceiling
  • kvm-worker-03 on CPUs 8-11 with a 1 GHz ceiling
  • host / emulator work on CPUs 0-1

What Was Confirmed​

  • virsh vcpupin matched the intended worker lanes
  • virsh emulatorpin kept emulator threads on 0-1
  • scaling_max_freq matched the intended per-lane caps
  • Grafana showed the expected frequency ceilings on the worker CPUs

Practical Read​

  • worker 1 got the fastest lane
  • worker 2 got the middle lane
  • worker 3 got the slowest lane
  • the host pair remained uncapped

That made the experiment easy to reason about:

  • kvm-worker-01 had the highest ceiling
  • kvm-worker-02 had a lower ceiling
  • kvm-worker-03 had the lowest ceiling

Why It Matters​

The clock-lane path proved that stakkr can enforce deterministic per-worker CPU ceilings on this host. That is useful for experiments and intentionally shaped node behavior, but it is a different model from the shared guest pool.