PCIe Intermediary for Low-Latency CPU-Network Packet Forwarding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

High packet forwarding delays occur when CPUs need to connect to network adapters in cloud computing environments, particularly in scenarios with multi-core processors and virtualization, leading to inefficiencies in data transmission.

Innovation Solution

Implementing an intermediate device, such as an FPGA or ASIC, that acts as a root node in the PCIe topology, enabling direct packet forwarding between CPUs and network adapters through SR-IOV technology, reducing the need for software conversions and utilizing an intermediate device to manage and forward packets directly.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a CPU needs to be connected to a network adapter through another CPU, then multi-core server configuration is achieved, but packet forwarding delay increases

Engineering Contradiction:
Improvepacket forwarding efficiencyVSAvoidpacket forwarding delay
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent introduces a PCIe switch as an intermediary device between CPUs and network adapters. The PCIe switch receives packets from one CPU and directly forwards them to the network adapter, bypassing the need for another CPU to handle the forwarding operation. This mediator enables direct packet routing while maintaining the multi-core server architecture, thus reducing forwarding delay without sacrificing productivity.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If virtualization technology is implemented, then resource pooling and dynamic scalability are achieved, but packet forwarding complexity increases

Engineering Contradiction:
Improveresource virtualization capabilityVSAvoidpacket forwarding complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments the packet forwarding function by separating virtualization management from physical packet routing. The PCIe switch handles physical packet forwarding independently using its own resource allocation mechanisms, while virtualization software manages virtual resource allocation. This segmentation allows virtualization to provide adaptability without adding complexity to the actual packet forwarding path, as the PCIe switch operates autonomously at the hardware level.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3813308B1Message forwarding method, intermediate device and computer device
Publication Date: 2025.08.06 HUAWEI TECH CO LTD
  • EP3813308B1 patent drawingFigure 1
  • EP3813308B1 patent drawingFigure 2A
  • EP3813308B1 patent drawingFigure 2B

AI summary

This application provides a packet forwarding method, an intermediate device, and a computer device. The computer device provided in this application includes a CPU, a network adapter, a bus, and an intermediate device, where the intermediate device is connected to both the CPU and the network adapter through the bus, and is configured to establish a correspondence between address information of an agent unit and address information of a function unit, and implement forwarding of a packet between the CPU and the network adapter based on the correspondence. The CPU can implement forwarding of a packet without using another CPU, so that operations such as a software conversion that are needed when the packet is forwarded to the another CPU are omitted, thereby reducing a delay in sending a data packet or a command packet. In addition, a problem of an excessively high delay because of insufficient bandwidth of an inter-CPU communications bus can be avoided.