Packet Processing Pipeline for NIC Policy Offload and Load Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data center operations face inefficiencies due to the use of CPU resources for network provisioning and storage, which could be better utilized for business logic and tenant hosting, and existing network interface devices lack efficient mechanisms for offloading packet processing operations.
Innovation Solution
Offload network policies and load balancing rules to a packet processing pipeline circuitry of a network interface device, which processes packets through a network policy table, recirculates the first packet to configure rules for the flow, and applies these rules to subsequent packets, thereby reducing CPU load and enhancing processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If CPU resources are used for network provisioning and storage operations, then networking and security functions can be performed, but CPU resources are consumed that could be used for business logic and tenant hosting
Solution Approach 1:
The patent extracts network provisioning, storage interface provisioning, load balancing, and security policy enforcement functions from the CPU to dedicated hardware components including network interface devices with programmable data planes and packet processing pipeline circuitry. This extraction allows CPU resources to be freed for business logic while maintaining networking and security functions through hardware-based processing.
2Reliability
If packet processing operations are handled by CPU, then network policies and load balancing can be applied, but processing efficiency is reduced due to CPU load
Solution Approach 1:
The patent introduces packet processing pipeline circuitry as an intermediary between network interface devices and the CPU. This pipeline includes stages for parsing packets, checking network policies, performing load balancing, and enforcing security rules. The pipeline operates independently of the CPU, providing accurate network policy application while maintaining high packet processing throughput through hardware-based parallel processing.
3Productivity
If network interface devices use fixed-function data planes, then simple packet forwarding can be performed, but flexibility for complex network policies and load balancing is limited
Solution Approach 1:
The patent implements programmable data planes in network interface devices that can be dynamically configured through a control plane. The data plane includes packet processing pipeline circuitry with multiple configurable stages that can be programmed to implement different network policies, load balancing algorithms, and security rules. This dynamic reconfigurability allows the system to adapt to changing requirements while maintaining high-speed hardware-based packet processing.
Data Source
AI summary
Examples described herein relate to a network interface device. In some examples, packet processing circuitry in the network interface device is to receive a first packet and based on the first packet being associated with an identifier for which an entry is not present in a look-up table accessible to the packet processing circuitry, the packet processing circuitry is to provide the identifier for the first packet and an action for the identifier of the first packet and cause the first packet to configure a second look-up-table accessible to the packet processing circuitry with the action for the identifier.