Halley: 광학 AI 데이터센터 네트워크를 위한 예측 제어 평면
이 논문은 대규모 AI 데이터센터의 광학 네트워크에서 발생하는 상관된 고장(correlated failure) 문제를 해결하기 위해 예측 기반 제어 평면(predictive control plane) 프레임워크인 Halley를 제안합니다. Halley는 레이저 풀 크기 조정(LightReserve), 스위치 스케줄링(FoldFit), 마이크로링 제어(ThermaLead), 위험 모듈 보호(TriageGuard)의 네 가지 구성 요소를 통해 각 계층의 다음 상태를 예측하고 고장 발생 전에 선제적으로 대응합니다. 이를 통해 고정된 예비 비율에서도 보장되던 가용성 한계를 극복하고, 각 계층에 대한 형식적 보증(formal guarantee)을 제공합니다.
A frontier AI model is trained in an AI data center, on tens of thousands of accelerators joined by a dedicated network, and every training step ends with all of them exchanging data at once. Between racks, that network is optical, and each of its four layers, from lasers to switches, runs a fixed or after-the-fact policy. Lasers sharing an enclosure, cooling, and power can fail together. At a fixed spare ratio, correlated failure provably caps a pool's availability however large it grows. In this paper, we present a comprehensive predictive control plane framework, called Halley, that reads each resource's next state and acts before the failure occurs. It has four components, each carrying a formal guarantee: LightReserve that sizes the laser pool to provide a coverage bound, FoldFit that schedules the switch within 11/9 of the optimal, ThermaLead that holds a microring with one-step-ahead control, and TriageGuard that protects the highest-risk modules first. Halley gives each layer a