Dynamic Node Routing for Multi-Processor Link Failures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large-scale multi-processor systems, fault-tolerance often degrades communication performance, and existing routing policies can lead to deadlocks when links fail, causing packet bottlenecks and system blockages.
Innovation Solution
A node routing method that learns the state of available links and reselects alternative links with different dimension sequence numbers to reroute packets, ensuring fault-tolerant communication by avoiding deadlocks through a routing policy controller with a state learning module and link selection module.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a routing policy controller formulates a routing policy to allow packets to be forwarded only by using the shortest route, then communication efficiency is improved, but when a link in the shortest route fails, packets wait for each other in a circular manner leading to deadlock
Solution Approach 1:
The routing policy is made dynamic by allowing nodes to switch between different routing paths based on link availability. Instead of being fixed to a single shortest route, the routing policy adapts when links fail by selecting alternative paths, thus resolving the contradiction between communication efficiency and fault-tolerance
Solution Approach 2:
The routing decision parameters are changed from static shortest-path only to dynamic path selection based on link state. By monitoring link availability and changing the routing parameters accordingly, the system maintains both efficiency and reliability
2Reliability
If fault-tolerance is implemented in multi-processor systems, then system reliability is improved, but communication performance is greatly degraded
Solution Approach 1:
The routing system performs self-adjustment by automatically detecting link failures and reconfiguring paths without external intervention. Each node monitors its own link state and independently selects alternative routes, maintaining communication performance while achieving fault-tolerance
Solution Approach 2:
The system pre-establishes multiple routing paths and maintains knowledge of link states in advance. When a failure occurs, the system already has alternative paths ready and can switch immediately, minimizing communication performance degradation while ensuring reliability
Data Source
AI summary
The present invention provides a node routing method of a multi-processor system, a controller and a multi-processor system. The method includes learning a state of an available link between nodes in the multi-processor system, where the multi-processor system includes a first subnet and the first subnet includes at least two connected nodes, and when at least one link in the first subnet fails, reselecting an available link between all nodes in the first subnet, so that the nodes in the first subnet use the reselected available link to route a packet, where the reselected available link is a link on each node in the first subnet except a link whose dimension sequence number is the same as that of the failed link, a dimension sequence number is numbers of a link at two end nodes, and numbers of a link at two end nodes are the same.