Scheduling Parallel Optical Circuit Switches for AI Training
This work provides a method to improve the efficiency and performance of AI training in datacenters by optimizing traffic scheduling on parallel optical circuit switches, which is a critical problem for large-scale AI infrastructure operators.
The paper addresses the challenge of efficiently scheduling time-varying AI traffic matrices across multiple parallel optical circuit switches, which are high-bandwidth and energy-efficient alternatives for AI fabrics. Their proposed algorithm, Spectra, decomposes traffic into permutations, schedules them across switches, and equalizes loads, resulting in a significant reduction in schedule makespan by an average factor of 1.4x on GPT AI workloads, 1.9x on MoE AI workloads, and 2.4x on standard benchmarks compared to state-of-the-art baselines.
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 $δ$. 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.