Rapid Forwarding Module for Bypassing CPU Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In content delivery networks (CDN), high CPU usage due to the inability to separate data and control planes leads to performance bottlenecks, especially under high concurrency, as all video data must be processed by the CPU, causing inefficiencies in data distribution.
Innovation Solution
Implementing a rapid forwarding module that receives data description information from the CPU, including address and length information, to directly read and forward user data to the network interface controller, bypassing the CPU and memory, allowing data transmission through an independent channel between the network interface controller and storage device without CPU intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If all video data is processed by the CPU and memory, then data distribution can be achieved, but CPU usage becomes high and performance bottleneck occurs under high concurrency
Solution Approach 1:
The patent segments the data distribution system into three independent functional modules: CPU (control plane), rapid forwarding module (data plane), and network interface controller. This segmentation allows the CPU to handle only control information while the rapid forwarding module handles data forwarding independently, resolving the contradiction by distributing workload across specialized components rather than concentrating all processing in the CPU.
Solution Approach 2:
The rapid forwarding module acts as an intermediary between the CPU and network interface controller. It receives data description information from the CPU, directly accesses storage devices to read video data, and forwards it to the network interface controller without CPU intervention. This intermediary resolves the contradiction by eliminating the CPU as a bottleneck in the data path while maintaining CPU control over the distribution process.
2Reliability
If CPU processes all data operations, then data distribution is achieved, but system performance is limited under high concurrency conditions
Solution Approach 1:
The system is divided into control functions (CPU managing data description information) and data functions (rapid forwarding module handling actual data transfer). This segmentation enables the rapid forwarding module to operate independently at high speed without being constrained by CPU processing capacity, thus improving system performance while maintaining reliability through preserved CPU control oversight.
Solution Approach 2:
The rapid forwarding module is designed to autonomously perform data reading from storage devices and forwarding to the network interface controller without requiring continuous CPU intervention. It self-manages the data plane operations based on control information received from the CPU, enabling high-concurrency handling and improving productivity while maintaining system reliability through independent error handling and control compliance.
3Ease of operation
If data passes through CPU and memory for all operations, then data can be distributed, but memory utilization increases and memory wall issue occurs
Solution Approach 1:
The rapid forwarding module serves as an intermediary that establishes a direct data path between storage devices and the network interface controller, bypassing the memory subsystem for data transfer operations. This intermediary enables simple data transmission operations while dramatically reducing memory utilization by eliminating the need for data to be loaded into and processed through system memory.
Solution Approach 2:
The patent extracts the data forwarding function from the traditional CPU-memory-network path and creates a separate rapid forwarding path that directly connects storage devices to the network interface controller. This extraction removes data transfer operations from the memory subsystem, reducing memory utilization while maintaining ease of operation through the standardized rapid forwarding module interface.
Data Source
AI summary
A method and an apparatus for rapid data distribution, the method includes: sending, by a central processing unit, data description information to a rapid forwarding module, where the data description information includes an address and length information of data requested by a user; reading, by the rapid forwarding module according to the data description information, the data requested by the user and forwarding the data requested by the user to a network interface controller; and sending, by the network interface controller, the data requested by the user to the user. By using the method provided in the present invention, after services are increased, only the network interface controller and a storage device need to be added, and cost for the memory and the central processing unit does not need to be increased.