Industry development and trend analysis of smart NICs under the background of cloud-network convergence

doi:10.12267/j.issn.2096-5931.2022.11.009

Abstract

Abstract:

Cloud-network convergence relies on the aggregation capabilities of the cloud to realize the collaboration, integration and innovation of multiple technical elements, and has become the best choice for the scenario-based needs of vertical industries. Massive data growth, frequent data interaction, and the slowdown of Moore’s Law all place higher requirements on the performance and cost of infrastructure, and smart NICs emerge as the times require. This paper compares the two acceleration processing modes centered on computing and data, analyzes the academic research progress of smart NICs, the characteristics and applications of mainstream products, and looks forward to the technological development trend of smart NICs. The analysis results show that the intelligent network card can maximize the overall performance of the system and “reduce the burden” for the hardware processor.

Key words: cloud-network convergence, smart NIC, hardware acceleration, heterogeneous computing

CLC Number:

TP393

ZHAO Jing, CHEN Yuanmou. Industry development and trend analysis of smart NICs under the background of cloud-network convergence[J]. Information and Communications Technology and Policy, 2022, 48(11): 69-78.

Add to citation manager EndNote|Ris|BibTeX

URL:

http://ictp.caict.ac.cn/EN/10.12267/j.issn.2096-5931.2022.11.009

http://ictp.caict.ac.cn/EN/Y2022/V48/I11/69

Figures/Tables 5

References 21

[1]	中国电信集团公司. 云网融合2030技术白皮书[R], 2020.
[2]	Dave Reinsel. IDC全球数据[EB/OL]. [2022-08-20]. https://www.idc.com/getdoc.jsp?containerId=prUS47560321.
[3]	IEEE P802.3bs. 200 Gbit/s and 400 Gbit/s ethernet task force[EB/OL]. [2022-08-20]. https://grouper.ieee.org/groups/802/3/bs/.
[4]	中国科学院计算技术研究所, 鄢贵海, 等. 专用数据处理器(DPU)技术白皮书[R], 2021.
[5]	TORK M, MAUDLEJ L, SILBERSTEIN M. Lynx: a smartNIC-driven accelerator-centric architecture for network servers[C]// ASPLOS’20:Architectural Support for Programming Languages and Operating Systems, 2020.
[6]	网络通信与安全紫金山实验室, 中国科学院计算技术研究所, 联通数字科技有限公司, 等. 未来网络白皮书——SmartNIC/DPU技术白皮书(2022版)[R], 2022.
[7]	GUO C, WU H, DENG Z, et al. RDMA over commodity ethernet at scale[C]. The 2016 Conference, 2016.
[8]	ZHU Y, MING Z, ERAN H, et al. Congestion control for large-scale RDMA deployments[J]. ACM SIGCOMM Computer Communication Review, 2015, 45(5):523-536. doi: 10.1145/2829988.2787484 URL
[9]	HU S, ZHU Y, PENG C, et al. Deadlocks in datacenter networks: why do they form, and how to avoid them[C]// The 15th ACM Workshop. ACM, 2016.
[10]	STEPHENS B, COX A L, SINGLA A, et al. Practical DCB for improved data center networks[C]// IEEE International Conference on Computer Communication (INFOCOM). IEEE, 2014.
[11]	MITTAL R, SHPINER A, PANDA A, et al. Revisiting network support for RDMA[C]// Conference on Special Interest Group on Data ComH munication(SIGCOMM), ACM, 2018: 313-326.
[12]	NI Z, WEI C, WOOD T, et al. A smart NIC-based load balancing and auto scaling framework for middlebox edge server[C]// 2021 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN). IEEE, 2021: 21-27.
[13]	HOLZINGER K, BIERSACK F, STUBBE H, et al. SmartNIC-basedload management and network health monitoring for time sensitive applications[C]// NOMS 2022-2022 IEEE/IFIP Network Operations and Management Symposium. IEEE, 2022: 1-6.
[14]	STEPHENS B, AKELLA A, SWIFT M. Loom: flexible and efficient {NIC} packet scheduling[C]// 16th USENIX Symposium on Networked Systems Design and Implementation (NSDI 19), 2019: 33-46.
[15]	GUO A, GENG T, ZHANG Y, et al. FCsN: a FPGA-centric smartNIC framework for neural networks[C]// 2022 IEEE 30th Annual International Symposium on Field-Programmable Custom Computing Machines (FCCM). IEEE, 2022: 1-2.
[16]	NVIDIA. NVIDIA BlueField数据处理器[EB/OL]. [2022-08-20]. https://www.nvidia.cn/networking/products/data-processing-unit/.
[17]	PATRICIA K. IPU:为云服务提供商量身打造的新型战略资源[EB/OL]. [2022-08-20].https://www.intel.cn/content/www/cn/zh/customer-spotlight/cases/ipu-new-strategic-resource-for-csp.html?wapkw=IPU.
[18]	赛迪顾问股份有限公司. 中国DPU行业发展白皮书[R], 2021.
[19]	金德路. CIPU面世,云计算将指向何方?[J]. 大数据时代, 2022(7):38-44.
[20]	王晓涛. 阿里云发布云数据中心专用处理器CIPU 替代CPU成为新管控加速中心[EB/OL]. (2022-06-15)[2022-08-20]. https://baijiahao.baidu.com/s?id=1735664036633957255&wfr=spider&for=pc.
[21]	张登科, 王兴伟, 贾杰, 等. 智能网卡研究新进展[J]. 小型微型计算机系统, 2022, 43(2):7.

主流厂商	代表性产品	核心处理器	应用方向	发布时间/年
阿里云	CIPU	FPGA+ASIC	基础设施资源快速云化、安全隔离、存储卸载等	2022
益思芯科技	Stargate	FPGA	网络、存储、计算的加速及卸载	2022
云脉芯联	metaFusion(DPU)	FPGA	虚拟化、网络加速、存储卸载、拥塞控制等	2022
NVIDIA	DPU(BlueField-3)	SoC: ARM+ASIC+ 专用加速器	网络加速、存储卸载、安全隔离、安全卸载等	2021
Marvell	OCTEON 10	SoC: ARM+ASIC	集成机器学习推理引擎、加密处理引擎、整形处理等	2021
芯启源	Agilio系列	NP-SoC	数据路径卸载、网络加速和卸载、加速密集计算	2021
腾讯	银杉	—	网络加速、存储加速	2021
Intel	IPU(FPGA IPU C5020X)	FPGA+X86 SoC	网络功能增强、基础设施任务加速等	2020
Fungible	F1 DPU	NP-SoC	全存储栈卸载、全安全栈卸载、大数据分析等	2020
Xilinx	Alveo U25	FPGA	网络、存储和计算加速	2020
Cisco	Nexus	FPGA	网络应用加速	2020
中科驭数	K1(DPU)	—	网络协议处理、数据加速、存储运算、安全加密等	2019
Broadcom	Stingray	SoC: ARM+ASIC	存储功能卸载	2018
华为	IN300	NP-SoC	实现高带宽高性能存储组网方案FCoE (Fiber Channel of Ethernet)	2018
Amazon	Nitro	—	虚拟化卸载、线速加密和解密	2017
Microsoft	Catapult v3	GP-SoC	面向深层神经网络加速	2017

主流厂商	代表性产品	核心处理器	应用方向	发布时间/年
阿里云	CIPU	FPGA+ASIC	基础设施资源快速云化、安全隔离、存储卸载等	2022
益思芯科技	Stargate	FPGA	网络、存储、计算的加速及卸载	2022
云脉芯联	metaFusion(DPU)	FPGA	虚拟化、网络加速、存储卸载、拥塞控制等	2022
NVIDIA	DPU(BlueField-3)	SoC: ARM+ASIC+ 专用加速器	网络加速、存储卸载、安全隔离、安全卸载等	2021
Marvell	OCTEON 10	SoC: ARM+ASIC	集成机器学习推理引擎、加密处理引擎、整形处理等	2021
芯启源	Agilio系列	NP-SoC	数据路径卸载、网络加速和卸载、加速密集计算	2021
腾讯	银杉	—	网络加速、存储加速	2021
Intel	IPU(FPGA IPU C5020X)	FPGA+X86 SoC	网络功能增强、基础设施任务加速等	2020
Fungible	F1 DPU	NP-SoC	全存储栈卸载、全安全栈卸载、大数据分析等	2020
Xilinx	Alveo U25	FPGA	网络、存储和计算加速	2020
Cisco	Nexus	FPGA	网络应用加速	2020
中科驭数	K1(DPU)	—	网络协议处理、数据加速、存储运算、安全加密等	2019
Broadcom	Stingray	SoC: ARM+ASIC	存储功能卸载	2018
华为	IN300	NP-SoC	实现高带宽高性能存储组网方案FCoE (Fiber Channel of Ethernet)	2018
Amazon	Nitro	—	虚拟化卸载、线速加密和解密	2017
Microsoft	Catapult v3	GP-SoC	面向深层神经网络加速	2017