Validation: Interference

Interference is the strongest research area in the project (Research Problem 2). This page is the GoMyRobotOS-side measurement methodology of the M3 Interference laboratory (GMR-INTERF v1); the physical experiment infrastructure is GoMyRobotBench.

The question

A partition does not interfere with another partition only through the CPU. It can interfere through:

CPU
cache
memory controller
interconnect
DMA
interrupts
I/O
accelerators

The methodology is: for each channel, stress one partition, measure the damage to another, and record which channels actually move the needle on which backend. The answer is different per backend by design, that delta is what Research Problem 1 (portable semantics) consumes.

The experimental design

Flight partition
       │
       │ latency / jitter
       ▼
    monitored
       ▲
       │
       │ stress
Service partition
 ├── CPU
 ├── cache
 ├── memory
 ├── DMA
 ├── IRQ
 └── I/O

Monitored partition: the contract’s timing_budget and declared communication endpoints define the baseline.
Stressor partition: one channel stressed at a time (then combined), with its own contract bounding its own resources, it stresses within its own contractual envelope, which is what makes the comparison fair and the result contractual.
Platform sweep: x86-64 → NG-ULTRA → HPSC, same conceptual partitions throughout (that invariance is deliberate: it is what makes results comparable across targets).

Measured quantities

Quantity	Interpretation for the contract
latency	window delay vs budget
jitter	variance of the window, the flight-relevant number
throughput degradation	endpoint rate vs `max_rate_hz` declaration
cache effects	which cache policies (if any) absorb the damage
memory bandwidth	can the declared memory regions hold their envelope?
interrupt latency	does `interrupts.owned` actually isolate interrupt delivery?
DMA interference	does `dma.permitted_regions` semantically extend to DMA timing?

Status

methodology:           specified (this page, M0)
GMR-INTERF v1:         Status: Planned (M3)
any interference data: none exists yet

Warning

No number on any GoMyRobotOS page is a measured interference result until an M3 (or later) artifact publishes it. “Low interference” and similar phrases are forbidden phrasings in this documentation at M0, this is an anti-claim rule as much as a methodology.