The Dual Horizon: A Rendezvous of Computing and Communication Services at the Optical Layer in Optical Computing-Communication Integrated Network

2025-05-24

AI Summary

In geographically distributed machine learning, frequent aggregation and computation of intermediate results impose substantial bandwidth pressure and lead to low resource utilization in traditional optical networks, which only support optical bypass. Method: This paper proposes an optical computing-communication integrated network architecture. It establishes, for the first time, a native optical-layer “computation-communication co-design” paradigm, transforming inherent interference in optical cross-connects into programmable optical computing resources. We formally define the joint routing, wavelength, and computation allocation problem and design a lightpath-level computing scheduling algorithm along with an extended RWA (Routing and Wavelength Assignment) modeling framework. Contribution/Results: Simulation on the COST239 topology demonstrates that, compared to conventional optical bypass, the proposed architecture achieves significant spectral efficiency gains and reduces bandwidth demand by 32.7%, validating the effectiveness and feasibility of native optical-domain computing for distributed training.

Technology Category

• Search and Optimization: Distributed Search
• Computer Vision: Learning & Optimization for CV
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

With the significant advancements in optical computing platforms recently capable of performing various primitive operations, a seamless integration of optical computing into very fabric of optical communication links is envisioned, paving the way for the advent of extit{optical computing-communication integrated network}, which provides computing services at the ligthpath scale, alongside the traditional high-capacity communication ones. This necessitates a paradigm shift in optical node architecture, moving away from the conventional optical-bypass design that avoids lightpath interference crossing the same node, toward leveraging such interference for computation. Such new computing capability at the optical layer appears to be a good match with the growing needs of geo-distributed machine learning, where the training of large-scale models and datasets spans geographically diverse nodes, and intermediate results require further aggregation/computation to produce the desired outcomes for the destination node. To address this potential use case, an illustrative example is presented, which highlights the merit of providing in-network optical computing services in comparison with the traditional optical-bypass mode in the context of distributed learning scenarios taking place at two source nodes, and partial results are then optically aggregated to the destination. We then formulate the new extit{routing, wavelength and computing assignment problem} arisen in serving computing requests, which could be considered as an extension of the traditional routing and wavelength assignment, that is used to accommodate the transmission requests. Simulation results performed on the realistic COST239 topology demonstrate the promising spectral efficiency gains achieved through the extit{optical computing-communication integrated network} compared to the optical-bypass model.

Problem

• Integrating optical computing into communication links for enhanced services
• Addressing routing and computing assignment in optical networks
• Improving spectral efficiency in distributed machine learning scenarios

Innovation

• Integrates optical computing into communication links
• Leverages lightpath interference for computation
• Formulates routing, wavelength, computing assignment problem