基于AWGR的全光学交换网络用于分布式机器学习

相关实验视频

Updated: May 30, 2025

Quasi-light Storage for Optical Data Packets

Quasi-light Storage for Optical Data Packets

Published on: February 6, 2014

Yuanzhi Guo, Xuwei Xue, Bingli Guo

|January 29, 2025

概括

我们开发了一个全光学交换网络,以加快分布式机器学习 (DML) 培训通信. 这种光学网络加速了DML培训,与传统电网相比,提高了成本和功率效率.

科学领域:

计算机科学计算机科学
电气工程电气工程
机器学习机器学习

背景情况:

在分布式机器学习 (DML) 中,训练数据集和模型的规模越来越大,使得主要的瓶从计算转移到通信.
现有的电气开关网络正在努力满足大规模DML日益增长的通信需求.

研究的目的:

提出和评估一个全光开关网络架构,旨在加快DML培训的通信阶段.
为了证明DML的光学切换的性能,成本和功率效率的好处.

主要方法:

实现一个全光学交换网络架构.
在高流量负载下对服务器对服务器通信延迟和错误率的实验验证 (0.9).
在拟议的架构上部署小型DML培训实验 (Resnet50,Resnet101,Vgg19),并与电气交换网络进行比较.

主要成果:

实现了无错误的数据包传输,在0.9.9的流量负载下,服务器对服务器的低延迟为385 ns.
与电气开关相比,对Resnet50,Resnet101和Vgg19的DML训练的演示加速为1.16x至1.48x.
显示了58.9%的成本效率提升和60.9%的功率效率提升,相比于三层脂肪树电气架构.

结论:

拟议的全光开关网络架构有效地解决了DML培训中的通信瓶.
光学开关在速度,成本和功率效率方面为大规模机器学习提供了显著的优势.
这项技术为更高效,更可扩展的分布式机器学习系统铺平了道路.

更多相关视频

Large Scale Energy Efficient Sensor Network Routing Using a Quantum Processor Unit

Large Scale Energy Efficient Sensor Network Routing Using a Quantum Processor Unit

Published on: September 8, 2023

Transmission of Multiple Signals through an Optical Fiber Using Wavefront Shaping

Transmission of Multiple Signals through an Optical Fiber Using Wavefront Shaping

Published on: March 20, 2017

相关概念视频

Ampere-Maxwell's Law: Problem-Solving

Ampere-Maxwell's Law: Problem-Solving

A parallel-plate capacitor with capacitance C, whose plates have area A and separation distance d, is connected to a resistor R and a battery of voltage V. The current starts to flow at t = 0. What is the displacement current between the capacitor plates at time t? From the properties of the capacitor, what is the corresponding real current?
To solve the problem, we can use the equations from the analysis of an RC circuit and Maxwell's version of Ampère's law.
For the first part of...

Network Function of a Circuit

Network Function of a Circuit

Frequency response analysis in electrical circuits provides vital insights into a circuit's behavior as the frequency of the input signal changes. The transfer function, a mathematical tool, is instrumental in understanding this behavior. It defines the relationship between phasor output and input and comes in four types: voltage gain, current gain, transfer impedance, and transfer admittance. The critical components of the transfer function are the poles and zeros.