– Assumptions: four PSI 3 MW engines on both sides, N+1 = 12 MW, N = 9 MW if one fails; 9 MW running load; a 2.7 MW batch arrives. – Control layer: tags batch work, caps it, and cuts the cap in 3 s when needed, so batch may sit above the line. – Result: 11.7 MW of racks instead of 9 on the same engines.
Elyrian
scheduler talks to the plant, caps on batch racks
engines oncap on batch racksbattery
live, must runbatch, taggedcap on batchN+1, engines onN, if one failsracks installedbattery out
Racks on the same four engines
Status quo
no information, no caps
engines oncap on racksbattery
load, untaggedbatch not admittedN+1, engines onN, if one failsracks installedbattery out
Racks on the same four engines
Model, not measurement. Four PSI 3 MW engines, all on, carrying 9 MW of live load on both sides. Batch visible 30 s ahead; the cap opens over 3 s so the engines follow it and the battery only covers the small mismatch; caps act within 1 to 3 s; battery 3 MW for a few seconds. Not shown: if an engine fails, the same cap is cut within 3 s, which is what allows batch above the line. The 30% holds only while everything above the line is tagged batch. Batch racks are shown empty before the job to isolate one batch; in practice batch fills them most of the time. Google runs its campuses on the same logic at 25% and above; the island version is ours to prove.