Dual P8/P32 cancellation scheduler
Routing latency-sensitive work to cancelable P32 and stable work to fast P8 can keep mixed-work overhead below 0.05%.
Diagnostic work, exact algebra and local capabilities are not treated as end-to-end mining advantage.
What was tested?
Routing latency-sensitive work to cancelable P32 and stable work to fast P8 can keep mixed-work overhead below 0.05%.
Why the test is meaningful
A dual scheduler can trade a small controlled throughput cost for bounded stale-work latency, but only if both routing and worst-case response gates hold.
mixed_overhead=1−throughput(25% P32+75% P8)/throughput(P8)gate: mixed_overhead≤0.0005all cancellation responses≤15 msHow it was tested
Validate exact routing and 2^32 work in each arm, run 48 equal-work measurements, eight cancellation trials and compute the registered 25% interruptible-mix cost.
What happened
Seven of eight cancellations met 15 ms; one reached 17.22 ms. P32 retention was 0.995006 (CI95 0.994708–0.995303), and modeled mixed overhead was 0.001253, above the 0.0005 gate.
Exactness and statistical controls
Routing and exactness passed, but both the worst-case latency and mixed-overhead technical gates failed.
What the result means
The proposed dual scheduler is not selected. Its policy does not justify deployment or changes to pool processes.
Limitations
- The 25% mix is a registered scenario, not every workload.
- One tail-latency failure is decisive under the all-trials rule.
- The result does not rule out materially different scheduling designs.
Evidence trail
Recreate exact P8/P32 routing, equal work and the 25% mix; retain the slowest cancellation and apply the fixed 0.0005 overhead gate.
Canonical variants
CANONICAL-EXP-168PREREGISTERED-CAMPAIGNSEALED-AUDITSource: internally audited canonical reports. Local filesystem structure, private headers and operational identifiers are excluded from publication.