.png){kind=icon}
.png){kind=icon}
Return
Design and Optimization of a Reconfigurable WSS-Based All-Optical Dragonfly Data Center Network
.png){kind=icon}
PRE
AI
C
E
Z
C
G
DOI：10.1109/jlt.2026.3699367.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The Dragonfly topology, known for its low diameter and large bi-section bandwidth, is particularly well-suited for Data Center Networks (DCNs) supporting large-scale artificial intelligence (AI) workloads. However, conventional Dragonfly-based DCNs rely on electrical switches, whose performance in terms of bandwidth, latency, and energy consumption is increasingly constrained by the physical limits of Moore's Law as AI workloads grow in scale and complexity. To overcome these limitations, optical circuit switching (OCS) technology has emerged as a promising alternative, offering ultra-high bandwidth, ultra-low latency, and significantly reduced energy consumption. In this paper, we propose a novel all-optical DCN architecture based on the Dragonfly topology, leveraging the flexible reconfiguration capabilities of Wavelength Selective Switches (WSSs). Building upon this architecture, we formulate and investigate the joint Routing, Wavelength Assignment, and Time Slot Scheduling (RWAT) problem. An Integer Linear Programming (ILP) model and three different heuristic algorithms are developed to address the RWAT problem. Simulation results across six typical service types show that the proposed heuristic algorithms can effectively solve the RWAT problem and achieve performance very close to the ILP optimal solution. Moreover, the proposed all-optical Dragonfly DCN can achieve the same switching scale with fewer WSSs than the all-optical Spine-Leaf architecture.
Keywords:
All-optical networks
data center networks
dragonfly
RWAT
wavelength selective switches
Journal
IF:
4.8
Papers:
1.7W
Citations:
3.8W
Organization
Cited Papers