Interconnection Network Topology Using Overlapping Bit Clusters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing interconnection networks face limitations in achieving high connectivity, low diameter, symmetry, and robustness due to cost and design complexity constraints, with traditional topologies like rings, meshes, and hypercubes failing to meet performance requirements for scalability and redundancy.
Innovation Solution
A method that associates n-bit binary identifiers with nodes, grouping them into overlapping sets to form fully connected clusters, allowing for customizable redundancy and path selection through minimal or non-minimal paths, resulting in a flexible and scalable network topology with increased node degrees and link counts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional network topologies (rings, meshes, hypercubes) are used, then network connectivity is limited, but cost and design complexity remain manageable
Solution Approach 1:
The patent segments the network topology into multiple levels of fully connected clusters. Each cluster is formed by grouping nodes based on overlapping sets of bit positions in their binary identifiers, creating a hierarchical structure where local clusters provide high connectivity and higher-level clusters provide long-range connections. This segmentation resolves the contradiction by achieving high connectivity through structured organization rather than random or traditional topologies.
Solution Approach 2:
The patent implements a nested cluster structure where clusters are formed at multiple levels. Lower-level clusters are nested within higher-level clusters, with each level providing additional connectivity. The overlapping set structure creates nested groups of nodes where nodes participate in multiple clusters simultaneously, achieving high connectivity while maintaining manageable complexity through the regular nested pattern.
2Reliability
If high node degree is used to achieve high connectivity and low diameter, then network robustness improves, but cost increases due to large number of links
Solution Approach 1:
The patent applies local quality by making different parts of the network have different connectivity characteristics. Nodes within the same cluster have high connectivity (fully connected), while nodes in different clusters have connectivity determined by the hierarchical level. This allows the network to achieve high robustness locally within clusters while controlling the overall number of links through the hierarchical structure, resolving the contradiction between robustness and link quantity.
3Speed
If traditional topologies are used, then design is simple, but network diameter is high and bisection bandwidth is low
Solution Approach 1:
The patent adds hierarchical dimensions to the network topology. Instead of using a single flat structure like traditional meshes or hypercubes, the invention creates multiple hierarchical levels of clusters. This dimensional approach allows nodes to reach each other through multiple pathways at different hierarchical levels, reducing network diameter and improving bisection bandwidth while maintaining a structured topology that is not overly complex.
4Adaptability or versatility
If network topology is made regular and scalable, then design complexity is reduced, but flexibility to meet specific redundancy requirements is limited
Solution Approach 1:
The patent introduces dynamics by allowing the overlapping sets to be configured differently to meet specific redundancy requirements. While the overall hierarchical cluster structure remains regular and scalable, the specific bit position groupings can be adjusted to provide varying levels of redundancy tailored to different workload requirements. This dynamic configurability resolves the contradiction between regularity and adaptability.
Data Source
AI summary
One embodiment of the present invention provides a system for generating an interconnection network. During operation, the system associates an n-bit binary identifier with each node in the interconnection network. The system also groups the n bits of each binary identifier into h overlapping sets a1, a2, . . . ah, wherein the sets a1, a2, . . . ah cover all of the n bits. For each set ai, the system forms fully connected clusters of nodes by creating a direct point-to-point link between any two nodes that have binary identifiers that differ in bits of the set a1 but are the same in other bit positions. Note that by varying the amount of overlap and the pattern of overlap between the overlapping sets, a1, a2, . . . ah, the interconnection network can be configured to accommodate different redundancy requirements.