A Unified Architecture for Accelerating Distributed DNN Training in Heterogeneous GPU/CPU Clusters.
Yimin JiangYibo ZhuChang LanBairen YiYong CuiChuanxiong GuoPublished in: OSDI (2020)
Keyphrases
- heterogeneous computing
- loosely coupled
- training process
- heterogeneous environments
- highly distributed
- heterogeneous systems
- data transfer
- real time
- distributed processing
- master slave
- distributed architecture
- graphics processing units
- multi agent architecture
- distributed systems
- graphics processors
- layered architecture
- graphic processing unit
- multithreading
- parallel computing
- heterogeneous data
- hierarchical architecture
- grid computing
- compute intensive
- distributed environment
- gpu implementation
- parallel processing
- computing platform
- distributed information systems
- scalable distributed
- commodity hardware
- distributed computing
- agent based architecture
- fuzzy clustering
- data points
- level parallelism
- processing units
- multi tier
- heterogeneous databases
- computing environments
- cluster analysis
- training set
- multi agent
- clustering algorithm
- web services
- high performance computing
- parallel algorithm
- management system
- data sources