Bipartite Mesh Network Routing via Virtual Lane Modification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-performance communication networks face challenges in preventing cyclic flow-control deadlocks and optimizing routing efficiency, particularly in Dragonfly topologies, where existing solutions do not effectively manage traffic without intermediate hosts and may lead to credit loops.
Innovation Solution
The network architecture employs a bipartite topology within groups and a mesh topology between groups, using virtual lane values to prevent cyclic flow-control deadlocks and implementing progressive Adaptive Routing (AR) by modifying packet virtual lane values to ensure direct routing without intermediate traversals, with nodes selecting routes based on virtual lane values and port information to prioritize direct paths over indirect ones.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If indirect adaptive routing is used to improve routing efficiency, then bandwidth utilization is enhanced, but cyclic flow-control deadlocks may occur
Solution Approach 1:
The patent changes the virtual lane parameter of packets when they traverse intermediate groups. Specifically, when a packet enters an intermediate group, its virtual lane value is modified (e.g., from VL0 to VL1), and when it exits, the virtual lane is changed again (e.g., from VL1 to VL2). This parameter transformation prevents cyclic deadlocks by ensuring that packets use different virtual lanes in different directions, breaking the symmetry required for credit loop formation while maintaining routing efficiency.
2Adaptability or versatility
If packets are routed via intermediate groups to optimize path selection, then routing flexibility is improved, but credit loops may be formed
Solution Approach 1:
The patent transforms the virtual lane parameter of packets traversing intermediate groups to prevent credit loops. When a packet enters an intermediate group from a source group, its virtual lane is changed (e.g., VL0→VL1). When it exits to a destination group, the virtual lane is changed again (e.g., VL1→VL2). This ensures that packets traveling through intermediate groups use different virtual lanes than direct packets, breaking the symmetry required for credit loop formation while preserving routing flexibility.
3Speed
If direct routing paths are prioritized to reduce latency, then speed is improved, but routing adaptability is reduced
Solution Approach 1:
The patent implements dynamic routing where nodes can adaptively select between direct and indirect paths based on current network conditions. The routing decision is not fixed but can change dynamically. Packets can take direct paths when available and optimal, or be routed through intermediate groups when beneficial, with the virtual lane modification enabling flexible path selection without creating deadlocks. This dynamic approach balances speed and adaptability.
Data Source
AI summary
A communication network includes multiple nodes, which are arranged in groups such that the nodes in each group are interconnected in a bipartite topology and the groups are interconnected in a mesh topology. The nodes are configured to convey traffic between source hosts and respective destination hosts by routing packets among the nodes on paths that do not traverse any intermediate hosts other than the source and destination hosts.