中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
首页 机构 成果 学者
  
  1. CAS IR Grid
  2. 计算技术研究所
  3. 中国科学院计算技术研究所
  4. 中国科学院计算技术研究所期刊论文
  5. 英文
Towards connection-scalable RNIC architecture

文献类型：期刊论文

作者Kang, Ning1,2; Wang, Zhan2; Yang, Fan2; Ma, Xiaoxiao1,2; Ma, Zhenlong1,2; Yuan, Guojun2; Tan, Guangming2
刊名JOURNAL OF SUPERCOMPUTING
出版日期2024-03-21
页码25
关键词Architecture design Network Interface Card (NIC) Remote Direct Memory Access (RDMA) Scalability problem
ISSN号0920-8542
DOI10.1007/s11227-024-05991-4
英文摘要Remote Direct Memory Access (RDMA) is a widely adopted optimization strategy in datacenter networking that surpasses traditional kernel-based TCP/IP networking through mechanisms such as kernel bypass and hardware offloading. However, RDMA also faces a scalability challenge with regard to connection management due to limited on-chip memory capacity in the RDMA Network Interface Card (RNIC). This necessitates the storage of connection context within RNIC's memory and induces considerable performance degradation when maintaining a large number of connections. In this paper, we propose a novel RNIC microarchitecture design that achieves peak performance and scales well with the number of connections. First, we model RNIC and identify two key factors that degrade performance when the number of connections grows large: head-of-line blocking when accessing the connection context and connection context dependency in transmission processing. To address the head-of-line blocking problem, we then combine a non-blocking connection requester and connection context management module to process prepared connections first, which achieves peak message rate when the number of connections grows large. Besides, to eliminate connection context dependency in RNIC, we deploy a latency-hiding connection context scheduling strategy, maintaining low latency when the number of connections increases. We implement and evaluate our design, demonstrating its successful maintenance of peak message rate (66.4 Mop/s) and low latency (3.89 mu s) while scaling to over 50,000 connections with less on-chip memory footprint.
资助项目National Key Research and Development Program of China[2021YFB0300700] ; International Partnership Program of Chinese Academy of Sciences[171111KYSB20180011] ; National Science and Technology Innovation Program 2030[2020AAA0104402] ; NSFC[61972380] ; National Key Research and Development Program[2021YFB0300700] ; Research and Innovation Program of State Key Laboratory of Computer Architecture, ICT, CAS[CARCH5407]
WOS研究方向Computer Science ; Engineering
语种英语
WOS记录号WOS:001190147700003
出版者SPRINGER
源URL[http://119.78.100.204/handle/2XEOYT63/38774]  
专题中国科学院计算技术研究所期刊论文_英文
通讯作者Wang, Zhan
作者单位1.Univ Chinese Acad Sci, Beijing 100089, Peoples R China
2.Chinese Acad Sci, Inst Comp Technol, Beijing 100089, Peoples R China
推荐引用方式
GB/T 7714		 Kang, Ning,Wang, Zhan,Yang, Fan,et al. Towards connection-scalable RNIC architecture[J]. JOURNAL OF SUPERCOMPUTING,2024:25.
APA Kang, Ning.,Wang, Zhan.,Yang, Fan.,Ma, Xiaoxiao.,Ma, Zhenlong.,...&Tan, Guangming.(2024).Towards connection-scalable RNIC architecture.JOURNAL OF SUPERCOMPUTING,25.
MLA Kang, Ning,et al."Towards connection-scalable RNIC architecture".JOURNAL OF SUPERCOMPUTING (2024):25.

入库方式： OAI收割

来源：计算技术研究所

浏览0
下载0
收藏0
其他版本

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。