Paper List of Workshops
OpticGAI: Generative AI-aided Deep Reinforcement Learning for Optical Networks Optimization
Siyuan Li, Xi Lin, Yaju Liu, Gaolei Li, Jianhua Li
Abstract: Deep Reinforcement Learning (DRL) is regarded as a promising tool for optical network optimization. However, the flexibility and efficiency of current DRL-based solutions for optical network optimization require further improvement. Currently, generative models have showcased their significant performance advantages across various domains. In this paper, we introduce OpticGAI, the AI-generated policy design paradigm for optical networks. In detail, it is implemented as a novel DRL framework that utilizes generative models to learn the optimal policy network. Furthermore, we assess the performance of OpticGAI on two NP-hard optical network problems, Routing and Wavelength Assignment (RWA) and dynamic Routing, Modulation, and Spectrum Allocation (RMSA), to show the feasibility of the AI-generated policy paradigm. Simulation results have shown that OpticGAI achieves the highest reward and the lowest blocking rate of both RWA and RMSA problems. OpticGAI poses a promising direction for future research on generative AI-enhanced flexible optical network optimization.
Rethinking Transport Protocols for Reconfigurable Data Centers: An Empirical Study
Federico De Marchi, Wei Bai, Jialong Li, Yiting Xia
Abstract: Fast-reconfiguring data center network architectures are rising as an answer to the expanding gap between traditional switch capacities and cloud traffic demands. These networks feature frequently changing connections and dynamic routing schemes, creating new challenges for the transport layer. We find that current available schemes have deficiencies when running on top of reconfigurable network fabrics, hindering their performance. Recent customized solutions are also unable to perform optimally when faced with the high number of topology configurations and routing strategies a reconfigurable network can boast. We conduct an empirical analysis of representative transport schemes for these architectures, which will provide insight for future transport designs.
Fast-tunable Graphene-based AWGR for Deep Learning Training Networks
Shicheng Zhang, Xuwei Xue, Bingli Guo, Yixuan Li, Wenzhe Li, Shikui Shen, Haoze Qian, Xiaojie Yin, Buzheng Wei, Guojun Yuan, Xiongyan Tang, Shanguo Huang
Abstract: Optical interconnection networks are often cited as ways to break through the energy-bandwidth limitations of conventional electrical wires and improve interconnect performance in data centers and high-performance computing (HPC). However, due to the rigorous demands on latency and bandwidth imposed by the trillions of parameters synchronization in the distributed training, current optical interconnect technologies, such as those based on Arrayed Waveguide Grating Routers (AWGR) or MEMS optical switches, have become critical performance bottlenecks in Deep Learning. In this paper, targeting the periodic patterns of internode interconnectivity and the multiple iterative feature in distributed training, we propose a graphene-based fast-tunable AWGR device. The switching characteristics of this device are highly compatible with the traffic features of distributed computing, which is expected to significantly improve the training efficiency of large-scale deep learning. Leveraging the strong electromagnetic wave confinement achieved with multilayer graphene, we achieve picosecond-level tuning speeds and center wavelength shifts across the full Free Spectral Range (FSR). Theoretical calculations and simulation results indicate that our optical interconnect approach speeds up the training times of deep learning applications around 33%.