Dorylus: Affordable, Scalable, and Accurate GNN Training with Distributed CPU Servers and Serverless Threads.
John ThorpeYifan QiaoJonathan EyolfsonShen TengGuanzhou HuZhihao JiaJinliang WeiKeval VoraRavi NetravaliMiryung KimGuoqing Harry XuPublished in: CoRR (2021)
Keyphrases
- scalable distributed
- distributed systems
- load balance
- training process
- multi threaded
- high accuracy
- lightweight
- data intensive
- training set
- data transfer
- wide area network
- high scalability
- high quality
- limited memory
- training phase
- central server
- computer networks
- distributed environment
- databases
- commodity hardware
- peer to peer
- real time
- learning algorithm
- remote server
- single point of failure
- distributed processing
- parallel computing
- file system
- steady state
- mobile agents
- training samples
- computationally efficient
- neural network