Snoop Multi-Cast Optimization in Mesh Network-on-Chip
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing processor architectures face challenges in efficiently managing cache coherency across multiple cores in a mesh topology, leading to increased latency and reduced bandwidth due to high overhead in snoop operations.
Innovation Solution
The implementation of an optimized snoop multi-cast with mesh regions, where the system-on-a-chip (SOC) is divided into regions with coherent tiles, and snoop vectors are generated and sent based on region IDs to minimize unnecessary traffic and optimize cache coherency operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional snoop operations are used across the entire mesh topology, then cache coherency is maintained, but network traffic and overhead increase significantly
Solution Approach 1:
The mesh topology is divided into multiple regions, and snoop operations are restricted to only those regions that contain caches with the same address tag as the accessed address. This segmentation prevents unnecessary snoop traffic from propagating across the entire mesh, reducing network overhead while maintaining cache coherency where needed.
2Reliability
If snoop operations are broadcast to all coherent tiles, then cache coherency is ensured, but latency increases due to traversing the entire mesh
Solution Approach 1:
The patent applies local quality by making snoop operation scope dynamic and location-dependent. Instead of uniform broadcasting to all tiles, the snoop is localized to specific regions based on the address tag match. This reduces the distance snoop operations must travel, thereby reducing latency while maintaining coherency in the relevant local regions.
3Adaptability or versatility
If mesh topology is used for processor interconnection, then scalability is improved, but snoop operation overhead increases with more cores
Solution Approach 1:
By segmenting the mesh into regions and using address tag-based filtering, the patent enables the snoop system to scale efficiently. As more cores are added to the mesh, the region division and tag-matching mechanism automatically limits snoop propagation to relevant subsets, preventing overhead from increasing linearly with the number of cores.
Data Source
AI summary
Processor data sharing is described. A system-on-a-chip (SOC) is accessed. The SOC includes a network-on-a-chip (NOC). The NOC includes an M×N mesh topology with a coherent tile at each point of the M×N mesh topology. The M×N mesh topology is divided into a plurality of regions. Each region in the plurality of regions includes one or more coherent tiles. A snoop operation is initiated by a first coherent tile within a first region. A snoop vector is generated by the first coherent tile for each region. The snoop vector for each region selects at least one other coherent tile. The snoop operation is sent by the first coherent tile for each region. The sending is based on the snoop vector for each region. The snoop operation is processed by the at least one other coherent tile.