Switchless Network Topology for Parallel Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As the number of nodes in parallel computation computer systems increases, the complexity and cost of network systems grow due to the need for multiple switches, leading to inefficient communication and increased complexity.
Innovation Solution
A switchless network topology system is implemented, where each node calculates and stores shortest paths using algorithms like Dijkstra's or Floyd-Warshall to create a routing table, allowing direct communication between nodes without relying on switches, thereby simplifying the network structure and maintaining cost efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If switches are used to accelerate communication between nodes, then communication speed is improved, but network structure complexity and cost increase
Solution Approach 1:
The patent extracts and eliminates switches from the network architecture by implementing a switchless topology where nodes directly communicate with each other. Each node computes shortest paths using algorithms like Dijkstra's or Floyd-Warshall and stores routing tables to enable direct communication, removing the need for intermediate switching devices and thereby reducing structural complexity while maintaining communication speed.
Solution Approach 2:
Each node in the network performs self-service by independently computing shortest paths to all other nodes and maintaining its own routing table. This distributed approach eliminates the need for centralized switch control, allowing nodes to autonomously determine communication routes and reducing overall network complexity.
2Power
If the number of nodes increases to enhance computational capability, then processing power is improved, but network communication complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-computing shortest paths between all pairs of nodes using algorithms like Dijkstra's or Floyd-Warshall before actual communication occurs. These pre-computed paths are stored in routing tables at each node, enabling O(1) route lookup during communication and avoiding the need for complex real-time path computation as nodes are added to the network.
Solution Approach 2:
The routing table mechanism serves multiple functions simultaneously: it stores shortest paths, enables direct communication routing, and provides path diversity options. This multi-functional approach handles communication complexity for any number of nodes uniformly, scaling well as computational capability increases.
3Productivity
If multiple layers of switches are used in fat tree structure, then communication efficiency is improved, but system cost increases
Solution Approach 1:
The patent removes switches from the fat tree structure entirely, replacing multi-layer switching infrastructure with direct node-to-node communication paths. This extraction eliminates the need for expensive switch hardware while maintaining communication efficiency through algorithmic optimization of routing paths.
Solution Approach 2:
The patent replaces the mechanical/swapping-based communication system (physical switches routing packets) with an algorithmic system (computational shortest path calculation and routing table lookup). This substitution eliminates hardware costs while maintaining or improving communication efficiency through software-based optimization.
Data Source
AI summary
A switchless network topology system for parallel computation is provided. The switchless network topology system includes a plurality of nodes, of which one node communicates with another node according to a routing table, where the creation of the routing table includes the following steps: calculating a shortest path of transmission paths between two nodes of the nodes, where the shortest path has no relay node or at least one relay node between the two nodes; removing the transmission path between a starting node and a first relay node of the shortest path and identifying another shortest path between the starting node and the destination node; repeating the step of removing and identifying to identify other shortest paths between the starting node and the destination node; and creating the routing table based on every shortest path between any starting node and any destination node of the nodes.