Multicast Routing for Mesh Network Bandwidth Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In multi-processor systems, especially in AI neural network processing, memory access bandwidth and network link bandwidth consumption are high due to individual processors accessing memory, and existing solutions like caching and multicast writing either fail to fully address the issue or increase programming complexity.
Innovation Solution
A multicast routing method for a mesh network system that configures different routing policies for interconnection devices based on their connection to external processing subsystems, balancing response data load and improving network performance by dynamically determining the shortest path and next-hop interconnection devices using distinct static routing policies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If static routing is used in the entire interconnection network, then routing is simple to implement, but some links may be congested and network performance is lowered
Solution Approach 1:
The patent segments the interconnection network into multiple processing subsystems, with each subsystem having its own response processing module. This segmentation allows different routing policies to be applied in different segments, balancing simplicity with performance optimization.
Solution Approach 2:
The patent introduces dynamic routing within processing subsystems by determining whether to use multicast routing or unicast routing based on the destination address. This dynamic approach allows the system to adapt to different traffic patterns while maintaining simple static routing between subsystems.
2Loss of energy
If multicast write technology is introduced, then transmission bandwidth consumption is reduced, but programming complexity is significantly increased
Solution Approach 1:
The patent introduces response processing modules as intermediaries between processors and memories. These modules automatically handle the complex multicast routing logic, shielding programmers from the complexity while achieving bandwidth optimization. The response processing modules act as mediators that translate simple processor requests into optimized multicast operations.
Solution Approach 2:
The response processing modules autonomously determine routing paths and distribute data to requesting processors without requiring programmer intervention. The system self-manages the multicast routing complexity, allowing programmers to use simple unicast-style programming while the system automatically optimizes transmission bandwidth usage.
3Productivity
If cache is added at memories, then memory bandwidth bottleneck is alleviated, but link bandwidth consumption in the entire network is not solved and implementation cost is high
Solution Approach 1:
Instead of adding caches at the memory level (vertical dimension), the patent introduces response processing modules at the network interconnection level (horizontal dimension). This dimensional shift allows the system to optimize link bandwidth consumption without requiring expensive memory caches, addressing the network-level bottleneck rather than the memory-level bottleneck.
Data Source
AI summary
A multicast routing method and an interconnection device for a mesh network system, a mesh network system and a configuration method thereof are provided. The method includes, at each internal interconnection device among multiple interconnection devices of each processing subsystem: in response to receiving a multicast access request to a destination memory, determining a shortest path from each internal interconnection device to the destination memory based on a topology structure of the mesh network system; where the internal interconnection device has no link connected to an external processing subsystem; in response to determining that the number of the shortest path is equal to one, routing the multicast access request to the destination memory along the shortest path; in response to determining that the number of the shortest path is greater than one, determining a next-hop interconnection device for the multicast access request based on a second static routing policy.