摘要
随着云计算向数据化智能化的方向演进,数据的流转与有效利用将为业务带来核心价值。大规模深度学习、机器训练等应用是极其依赖算力的,大量的信息交互对网络提出了很高的要求,由此需要一个低时延、无丢包、高吞吐的算力网络。考察RDMA[1]技术在数据中心中的应用,并分析其对于未来云数据中心高性能集群计算的影响。
As cloud computing evolves towards data and intelligence,the flow and effective use of data will bring core value to the business.Applications such as large-scale deep learning and machine training are extremely dependent on computing power.A large number of information interactions place high demands on the network,which requires a low latency,no packet loss,and high throughput computing network.This paper mainly examined the application of RDMA[1]technology in data centers and analyzed its impact on future high-performance cluster computing in cloud data centers.
作者
涂晓军
孙权
蔡立志
Tu Xiaojun;Sun Quan;Cai Lizhi(China UnionPay Co.,Ltd.,Shanghai 201201,China;Fudan University,Shanghai 200433,China;Shanghai Key Laboratory of Computer Software Testing&Evaluating,Shanghai Development Center of Computer Software Technology,Shanghai 201112,China)
出处
《计算机应用与软件》
北大核心
2021年第3期22-25,45,共5页
Computer Applications and Software
基金
上海市优秀学术/技术带头人计划项目(19XD1433700)。
关键词
RDMA
低延时
大算力网络
RDMA
Low latency
High throughput computing network
