Litcius/Paper detail

Prague: High-Performance Heterogeneity-Aware Asynchronous Decentralized Training

Qinyi Luo, Jiaao He, Youwei Zhuo, Xuehai Qian

202073 citationsDOIOpen Access PDF

Abstract

Distributed deep learning training usually adopts All-Reduce as the synchronization mechanism for data parallel algorithms due to its high performance in homogeneous environment. However, its performance is bounded by the slowest worker among all workers. For this reason, it is significantly slower in heterogeneous settings. AD-PSGD, a newly proposed synchronization method which provides numerically fast convergence and heterogeneity tolerance, suffers from deadlock issues and high synchronization overhead. Is it possible to get the best of both worlds --- designing a distributed training method that has both high performance like All-Reduce in homogeneous environment and good heterogeneity tolerance like AD-PSGD?

Topics & Concepts

Synchronization (alternating current)Computer scienceAsynchronous communicationHomogeneousDistributed computingConvergence (economics)Overhead (engineering)DeadlockData synchronizationTraining (meteorology)Bounded functionComputer networkOperating systemEconomic growthWireless sensor networkMeteorologyPhysicsThermodynamicsMathematicsEconomicsMathematical analysisChannel (broadcasting)Cloud Computing and Resource ManagementGraph Theory and AlgorithmsDistributed and Parallel Computing Systems
Prague: High-Performance Heterogeneity-Aware Asynchronous Decentralized Training | Litcius