Scheduling Parallel Optical Circuit Switches for AI Training

TL;DR

This paper introduces Spectra, a scheduling algorithm for parallel optical switches in AI training datacenters, significantly reducing scheduling time and approaching optimal bounds by efficiently managing traffic matrices with reconfiguration delays.

Contribution

Spectra is a novel three-step scheduling algorithm that decomposes traffic matrices, assigns permutations across switches, and balances loads, outperforming existing methods in AI training workloads.

Findings

Reduces schedule makespan by up to 2.4x on benchmarks.

Achieves near-optimal makespans close to theoretical lower bounds.

Outperforms state-of-the-art algorithms on realistic AI workloads.

Abstract

The rapid growth of AI training has dramatically increased datacenter traffic demand and energy consumption, which has motivated renewed interest in optical circuit switches (OCSes) as a high-bandwidth, energy-efficient alternative for AI fabrics. Deploying multiple parallel OCSes is a leading alternative. However, efficiently scheduling time-varying traffic matrices across parallel optical switches with non-negligible reconfiguration delays remains an open challenge. We consider the problem of scheduling a single AI traffic demand matrix $D$ over $s$ parallel OCSes while minimizing the makespan under reconfiguration delay $\delta$. Our algorithm Spectra relies on a three-step approach: Decompose $D$ into a minimal set of weighted permutations; Schedule these permutations across parallel switches using load-aware assignment; then Equalize the imbalanced loads on the switches via controlled permutation splitting. Evaluated on realistic AI training workloads (GPT model and Qwen MoE expert routing) as well as standard benchmarks, Spectra vastly outperforms a baseline based on state-of-the-art algorithms, reducing schedule makespan by an average factor of $1.4\times$ on GPT AI workloads, $1.9\times$ on MoE AI workloads, and $2.4\times$ on standard benchmarks. Further, the makespans achieved by Spectra consistently approach newly derived lower bounds.