Six requirements for integrating SmartNIC/DPU into existing network infrastructure

Six requirements for integrating SmartNIC/DPU into existing network infrastructure

The CPU has long been considered the backbone of computing processing devices, while the GPU is used to handle real-time graphics. However, it is the data processing unit - DPU - that has recently gained attention.

SmartNIC/DPU

Network interface cards (NICs) have come a long way since the late 1980s. However, while they were advanced for their time, they were limited to simple connections between two computers or on a local network.

SmartNIC is a programmable network card that includes additional computing resources beyond conventional NIC. The programmable resources of SmartNIC can be composed of Arm cores, network processing units (NPUs), and other programmable units.

Many see DPUs as the next step in the evolution of SmartNICs.

DPUs are programmable devices that can assist in data processing through hardware acceleration. SmartNICs effectively connect servers to the network, and their programmable components are equipped with accelerators that can efficiently process data, thereby freeing up server space for application processing, but are not suitable for multitasking. Although SmartNICs can do the work of DPUs to some extent, their composition and functions are different.

SmartNIC provides extensive data storage and programmability, but it requires DPU to provide more virtualization capabilities. SmartNIC manages data with a certain upper limit, but when used in conjunction with DPU, it can break through data limitations, thus creating reliable programmable technology and bringing higher network processing capabilities.

In many cases, the concepts of SmartNIC and DPU are often used interchangeably.

According to data from research institutions, from 2019 to 2022, domestic and foreign companies have applied for numerous patents involving DPUs and related smart network cards, with Intel ranking first, Huawei ranking second, and Google ranking third. Chip and processor manufacturers such as AMD and Nvidia have committed to developing and manufacturing DPUs. Asus, Dell, and AMD have integrated with systems using Nvidia DPUs. AWS is also a major user of DPUs.

Why do existing network facilities need to integrate SmartNIC/DPU?

Overloaded Servers

The continued pressure of Moore's Law drives the miniaturization of semiconductor technology every two years. The development of quantum tunneling, reduction in loss layer size, and parasitic capacitance has led to interconnect bottlenecks in server integrated circuits. The reduction in hardware size has driven faster server expansion and increased the overall cost of the CPU.

Storage and delivery

With predictions pointing to the Gigabyte Era by 2040, network infrastructure is overwhelmed with the need to store data and deliver it on demand. To reduce the burden on servers, SmartNIC/DPU infrastructure provides data centers with gigabytes of additional storage, freeing up enough CPU cycles to perform high-value enterprise tasks and store critical information.

Power consumption

Increasing server workloads are associated with additional hardware and equipment required for power management. This hardware consumes additional power, increasing the cost of network infrastructure management. Power distribution units, backups, and cooling mechanisms can help to some extent, but SmartNIC/DPU infrastructure directly does not require the deployment of additional hardware, and the total power consumption decreases as the number of cores decreases.

Defense against attacks

SmartNIC/DPU infrastructure is able to perform network security functions at the hardware level, including packet packaging, hashing, compression and decompression, encryption and decryption, cryptographic algorithms, firewalls, and DDoS attack defense. SmartNIC/DPU can handle security operations and eliminate any threats without interfering with the server. The latest SmartNIC/DPU has firewalls and advanced DDoS attack defense mechanisms to support distributed security systems.

Requirements for integrated SmartNIC/DPU infrastructure

SmartNIC/DPU can meet the growing demand for high performance, low latency and secure data transmission. But the implementation of SmartNIC/DPU requires a reliable network infrastructure. The following are some important requirements for integrating SmartNIC/DPU infrastructure.

High connectivity

NICs are typically used in low-bandwidth network infrastructures with speeds in the 10 Gbps range. SmartNIC/DPU infrastructure is more suitable for large enterprises that support large bandwidths, handle high network traffic, and network speeds exceeding 100 Gbps, with socket acceleration through multiple network layers to offload and save server cycles for application-specific high-value tasks.

Ultra-low latency

SmartNIC/DPU implementations perform best in wired infrastructures where high-speed computations need to be performed using InfiniBand SmartNIC/DPU infrastructure. In such networks, Remote Direct Memory Access (RDMA) allows direct access to the memory of another computer, providing ultra-low latency without involving the kernel.

Traditional infrastructure is better suited for Ethernet and faces complexity when implementing SmartNIC/DPU. RDMA over Converged Ethernet supports Ethernet-based SmartNIC/DPU infrastructure, providing low latency and higher throughput.

Complex network architecture

SmartNIC/DPU is suitable for the server side of the client-server model of enterprises and large cloud providers. Network architectures such as software-defined networking (SDN), network function virtualization (NVF), high-speed computing and complex infrastructure provide conditions for SmartNIC/DPU to achieve hardware acceleration.

Distributed cloud computing companies from industries such as telecommunications, SaaS, e-commerce, and streaming platforms have deployed hundreds or thousands of SmartNICs for acceleration. These large organizations have network architectures that deliver microservices and process big data.

Network switch compatibility

A network can use multiple types of network switches. The compatibility and configurability of SmartNIC/DPU with existing hardware (e.g., switches, operating systems) in the network is critical.

Some SmartNICs/DPUs may not be compatible with the network infrastructure, requiring the selection of compatible SmartNICs/DPUs for existing network switches, including interfaces, protocols, programmability, configuration, and use cases.

SDN Integration

Large enterprises often use SDN to connect, manage and accelerate their workflows across multiple workplaces around the world. Integrating SmartNIC/DPU infrastructure in such networks can relieve the SDN centralized controller from the task of effectively managing the network.

Because SmartNICs/DPUs are programmable, they can accelerate data plane and application-specific tasks in routers and switches in SDN, including traffic forwarding, packet filtering, encryption and decryption, firewalls, etc.

NFV Features

SmartNIC/DPUs are also used in networks that adopt NFV. NFV networks require the implementation of SmartNIC/DPU infrastructure because multiple virtual machines run in a shared physical infrastructure, eliminating the need for dedicated hardware.

This is similar to deploying virtual microservers in a data network, each with its own resources. Examples of NVF acceleration include virtual extensible LAN, micro-segmentation, and load balancing.

After implementing SmartNIC/DPU in the network infrastructure, the network team can turn its focus to managing SmartNIC/DPU. Regular quality checks of SmartNIC/DPU infrastructure after deployment are key to continuous acceleration of intelligence. SmartNIC/DPU can conduct regular reviews of network, storage and security functions themselves. Effective network management ensures seamless performance of SmartNIC/DPU through monitoring software and key indicators.

Monitoring tools continuously collect and analyze data such as latency, speed, throughput, and packet loss. Once a tool or SmartNIC/DPU detects an error, it must be automatically troubleshooted to get a solution. In addition to troubleshooting, network management also ensures regular updates of SmartNIC/DPU infrastructure to meet the latest technology requirements. In addition, the network orchestration platform can also automatically configure SmartNIC/DPU to comply with current security operation regulations and standards.

summary

Integrating SmartNIC/DPU with existing network facilities is the key to ensuring its successful application in actual production environments. This seamless integration can improve overall performance and efficiency, reduce host processor load, and support network virtualization, bringing greater flexibility and manageability to enterprises. At the same time, the advanced network functions of SmartNIC/DPU can also enhance network security and protect enterprise data and assets.

As technology continues to advance and demand continues to change, SmartNIC/DPU will continue to play a key role in contributing to the continuous optimization of network performance and the improvement of application processing. We can foresee that the future network will be more intelligent and efficient, and bring more innovative possibilities to enterprises.

<<:  Google says TCP congestion control algorithm BBRv3 performs well and will be submitted to the Linux kernel mainline this month

>>:  Six key advantages of optical interconnect technology

Recommend

Omdia: Global Gigabit Broadband Users to Exceed 187 Million by 2025

LightReading reports that the COVID-19 pandemic h...

JuHost: Hong Kong VPS 40% off from $2.99/month, 1GB/20GB SSD/1TB@100Mbps

JuHost was registered in Hong Kong in early 2023,...

HostNOC: $39.99/month-E3-1230v2/16GB/1TB/100TB/Los Angeles data center

HostNOC is a foreign hosting company founded in 2...

Talk about the past and present of programmable network elements

[[402793]] This article is reprinted from the WeC...

Comprehensive understanding of TCP/IP knowledge system structure summary

1. TCP Knowledge System We analyze the TCP knowle...

...