M3 Interference Laboratory

Objective

Build the GMR-INTERF v1 interference laboratory and make interference measurements reproducible, the gate that converts interference from folklore into data for the project.

Scope

  • stressors per channel: CPU, cache, memory, DMA, IRQ, I/O

  • metrics per run: latency, jitter, throughput, interference

  • platform progression:

x86-64
   ↓
NG-ULTRA
   ↓
HPSC        (as M4 research permits)

  • the GoMyRobotOS-side methodology is [validation/interference] (../validation/interference); the physical experiment infrastructure is GoMyRobotBench ([components/gomyrobotbench] (../components/gomyrobotbench))

Deliverables

  • GMR-INTERF v1 (stress recipes + metric collection)

  • per-channel interference data on x86-64 (first), NG-ULTRA (second)

  • dataset documentation: stress recipe names, monitored partition (contract-identical flight partition), and measured quantities

  • findings on which interference channels matter per backend (RP2 feed)

Tests / acceptance gate

Interference measurements are reproducible.

Concretely: stress recipe → monitored run → metric set, rerunnable by two independent operators and producing values within a documented tolerance. Without reproducibility the whole cross-target comparison in RP1/RP2 is unfounded, so this is why this gate is procedural, not numeric.

Current status

Planned (after M1/M2 baseline existence).

Known limitations

  • M3 measures interference, not safety: a measured jitter is an engineering number, not a compliance certificate.

  • “Measured on x86-64” ≠ “measured on NG-ULTRA”: each platform’s data lands when that platform’s run exists; the sweep is progressive, and any page that reports numbers must name the platform it was taken on.

  • HPSC numbers, if any, are research-track and labeled as such (M4).