Research — Datacenter Controls · Live
Redundancy and failover math that scores the candidates you supply — it never invents interlock logic.
A compute-only library for the redundancy and control layer above power and cooling design: availability for N, N+1 and 2N-style topologies, power-path and water-path failover timing margins, transformer capacity planning and valve actuator torque. It scores failover sequences the caller supplies and never decides pass/fail.
Interactive demos
- K-of-N availability →Choose N components, how many must be up, and each one’s availability to see system availability, a nines gauge and derived downtime per year, with the independence assumption printed on every result. Draft, unvalidated.Auto ▶
- Failover race →Three caller-supplied failover sequences race a time budget and a required capacity; retime any step. The function scores candidates and never generates one.Auto ▶
See harnesses →the toolsets that sit on top of these functions.
What it can do
calculate_redundant_system_availabilityK-of-N availability from per-component availabilities; discloses its independence assumption.
calculate_component_availabilityAvailability of a single component.
calculate_transfer_switch_ride_through_marginMargin for power transfer-switch ride-through timing.
evaluate_failover_sequence_candidatesScores caller-supplied failover sequences against a time budget and capacity; reports the fastest passing one.
calculate_valve_actuator_torqueRequired actuator torque from differential pressure, diameter and a safety factor.
calculate_transformer_capacity_planAdditional fixed-size transformer units needed for a rack load, with surplus or shortfall.
A worked case
Input
Two components at 0.995 availability each, run in parallel (one required).
Result
System availability 0.999975. Required in series (both needed), it drops to 0.990025. A valve at 500 kPa differential, 0.1 m diameter, coefficient 0.15, safety factor 1.25 needs 93.75 N·m.
Where this stands
Draft; no reliability or controls engineer has reviewed it yet. The availability math is cross-checked against closed-form series, parallel and 2-of-3 cases, and an AI-assisted pre-review found and fixed three issues. IEEE 493 and ISA-75.19.01 were not read directly.
Open questions
- The independence assumption is unvalidated against real common-cause failures such as shared feeds and cooling loops, so results may be optimistic.
- The torque coefficient is caller-supplied with no plausibility guard.