DLTAP: A Network-efficient Scheduling Method for Distributed Deep Learning Workload in Containerized Cluster Environment

dyweb / papers-notebook

:page_facing_up: :cn: :page_with_curl: 论文阅读笔记（分布式系统、虚拟化、机器学习）Papers Notebook (Distributed System, Virtualization, Machine Learning)

Apache License 2.0

2.14k stars 250 forks source link

Open yylin1 opened 5 years ago

yylin1 commented 5 years ago

yylin1 commented 5 years ago

提出了一種稱為Deep Learning Task Allocation Priority （DLTAP）排成方法，主要目標以排成決策都考慮ps任務和工作任務的聚合度，特別是減少跨節點網絡傳輸流量，以致減少DNN訓練時間。