https://github.com/MachineLearningSystem/omnireduce

https://github.com/MachineLearningSystem/omnireduce

Last synced: 9 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/MachineLearningSystem/omnireduce
• Owner: MachineLearningSystem
• Fork: true (sands-lab/omnireduce)
• Created: 2021-12-11T10:56:29.000Z (almost 3 years ago)
• Default Branch: master
• Last Pushed: 2021-08-15T14:09:29.000Z (over 3 years ago)
• Last Synced: 2024-10-20T18:34:52.285Z (27 days ago)
• Size: 112 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-AI-system - Efficient Sparse Collective Communication and its application to Accelerate Distributed Deep Learning SIGCOMM'21

README

        
# OmniReduce

OmniReduce is an efficient sparse collective communication library. It maximizes effective bandwidth use by exploiting the sparsity of data.

For clusters without RDMA support, OmniReduce uses Intel DPDK for kernel bypass. GPUDirect can also be used where available.

## Contents

- omnireduce-DPDK: source code of DPDK-based OmniReduce

- omnireduce-RDMA: source code of RDMA-based OmniReduce

- [experiments](https://github.com/sands-lab/omnireduce-experiments): micro-benchmark and end-to-end scripts

## Publications

[OmniReduce](https://sands.kaust.edu.sa/project/omnireduce/) accepted at SIGCOMM’21.