ATRO: A Fast Solver-Free Algorithm for Topology and Routing Optimization of Reconfigurable Datacenter Networks

2025-07-18

AI Summary

Joint topology and routing optimization in reconfigurable datacenter networks (DCNs) is computationally challenging due to strong coupling and scalability constraints. Method: This paper proposes ATRO, a solver-free, fast alternating optimization framework. It decouples the problem into topology and routing subproblems, solved iteratively: the topology subproblem leverages monotonicity to design an Accelerated Binary Search Method (ABSM), guaranteeing strict reduction of the maximum link utilization per iteration; the routing subproblem integrates state-of-the-art traffic engineering accelerators. Contribution/Results: ATRO converges to the global optimum in single-hop DCNs and significantly outperforms baseline methods in multi-hop settings—achieving high solution quality, millisecond-scale runtime, and excellent scalability. It establishes a practical, efficient optimization paradigm for large-scale, dynamic DCNs.

Technology Category

• Search and Optimization: Combinatorial Optimization
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• 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

Abstract

The growing scale and complexity of reconfigurable data center networks (DCNs) demand more scalable and efficient algorithms for computing logical topologies and routing. Reconfigurable DCNs typically operate in two modes: one-hop configurations that require frequent topology optimization (TO), and multi-hop scenarios that involve joint topology and routing optimization (TRO). In both cases, the combinatorial nature of topology decisions makes it difficult for existing methods to balance solution quality and runtime efficiency. To address this, we introduce Alternating Topology and Routing Optimization (ATRO), a solver-free framework that alternates between TO and routing optimization (RO). This decomposition exploits two key insights: first, each alternating update step monotonically reduces maximum link utilization (MLU), ensuring consistent performance improvement across iterations; second, the TO subproblem, equivalent to one-hop optimization, exhibits a monotonic structure that enables optimal solutions via an efficient Accelerated Binary Search Method (ABSM). To preserve the solver-free design, RO is solved using existing Traffic Engineering accelerators. ATRO attains the global optimum in one-hop scenarios and significantly outperforms baselines in multi-hop settings in terms of both runtime and solution quality. Evaluations confirm its scalability and robustness across diverse DCNs.

Problem

• Scalable algorithms for reconfigurable DCN topology and routing
• Balancing solution quality and runtime efficiency in DCN optimization
• Efficient solver-free framework for joint topology and routing optimization

Innovation

• Solver-free alternating TO and RO framework
• Accelerated Binary Search Method for TO
• Traffic Engineering accelerators for RO