On Moving Warm Air Instead of Cold Water [/ɔn ˈmuvɪŋ ˈwɔrm ˈɛr ˌɪnˈstɛd əv ˈkoʊld ˈwɔtər/] n - The idea. In one of the older halls, we essayed to use warm aisle exhaust air to pre-heat the return side of a small liquid-cooling loop. The theory was plain: why cast away heat at 35 °C when it might be carried into water bound for the chiller in any case? The hardware was cheap, a fan coil, a pump, some ducting, and the energy model looked favorable.
What happened. [/wɒt ˈhæpənd/] n - The system served in steady state; in transition it was not elegant. Each time the IT load stepped, as happens often in a training cluster, the air temperature lagged the water temperature by a minute or two. By that lag the pre-heat loop went on adding heat when the load had just fallen, and drew back when the load had just spiked. The control loop spent most of its days chasing the wrong quarry.
Why we stopped. [/waɪ wiː ˈstɒpt/] n - The savings were real but fragile. The operators could not trust the loop to hold steady through ordinary variation, and the manual interventions it demanded consumed more attention than it was worth. It also muddied the room’s response to a true chiller failure, for the pre-heat loop added a second, slower thermal path.
What we kept. [/wɒt wiː ˈkɛpt/] n - The experiment taught us how much control latency matters in Direct-to-chip liquid cooling loops. The notion of heat reuse migrated to a central heat-recovery plant, where the volumes are larger and the transients slower, which suits the concept far better.