Scheduling Parallel Optical Circuit Switches for AI Training

2026-03-07

AI Summary

This work addresses the traffic scheduling problem on parallel optical circuit switches with non-zero reconfiguration delay, aiming to minimize makespan under the high-bandwidth and low-energy demands of AI training. The authors propose a three-stage scheduling approach: first decomposing the traffic matrix into weighted permutations, then assigning these permutations to parallel optical switches in a load-aware manner, and finally applying a controlled permutation splitting strategy to balance switch loads. This method is the first to integrate permutation decomposition with load balancing while accounting for reconfiguration overhead, achieving performance that closely approaches a newly derived theoretical lower bound. Experimental results demonstrate significant makespan reductions—by factors of 1.4× and 1.9× on representative AI workloads such as GPT and Qwen MoE, respectively—and up to 2.4× on standard benchmarks.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Machine Learning: Hardware-aware ML
• Search and Optimization: Algorithm Configuration

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

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.

Problem

• optical circuit switches
• AI training
• traffic scheduling
• reconfiguration delay
• parallel switches

Innovation

• optical circuit switching
• AI training traffic scheduling
• parallel OCS
• reconfiguration delay
• makespan minimization