Forwarding Chip Link Fault Handling for AI Training Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing link fault handling in network devices during the training of large-scale AI models is inefficient due to lengthy fault notification and route recalculation times, leading to prolonged training processes and potential errors.
Innovation Solution
A method and apparatus for processing link faults in forwarding chips of network devices that enable timely route switching by directly setting forwarding table entries to an invalid status based on status information from connected network devices, without involving the control plane, using packets with device IDs and status information to quickly identify and disable faulty links.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the control plane recalculates routes and modifies forwarding table entries after receiving route update messages, then the network can avoid communicating over failed links, but the fault notification and route recalculation take a long time, leading to untimely fault convergence
Solution Approach 1:
The patent extracts the forwarding table entry modification function from the control plane and places it directly in the forwarding chip. When a link fault occurs, the forwarding chip can directly set the corresponding forwarding table entry to invalid without waiting for control plane recalculation, achieving immediate fault isolation while maintaining network reliability
Solution Approach 2:
The patent introduces a status information field in forwarding table entries as an intermediary mechanism. This field allows the forwarding chip to quickly mark entries as invalid when link faults occur, enabling fast fault convergence without complex control plane intervention while maintaining proper routing avoidance
2Productivity
If the forwarding chip directly sets forwarding table entries to invalid status based on status information from connected network devices, then route switching becomes faster, but the system requires additional mechanisms to ensure proper fault notification and entry validation
Solution Approach 1:
The patent pre-loads forwarding table entries with a status information field that can be quickly set to invalid. This preliminary setup allows the forwarding chip to immediately disable faulty routes without complex real-time calculations, achieving fast route switching while keeping the mechanism simple
Solution Approach 2:
The patent implements a feedback mechanism where the forwarding chip monitors link status through status information and automatically adjusts forwarding table entries. This closed-loop feedback enables automatic fault response with simple logic: when status indicates failure, set entry to invalid; when status recovers, restore entry
Data Source
Figure 1~2
Figure 3~6
Figure 7~9
AI summary
The embodiments of the present application relate to the field of communications. Provided are a link fault handling method and apparatus, and an electronic device and a medium. The technical solution in the embodiments of the present application comprises: a forwarding chip of a first network device receiving a first message sent by a second network device, wherein the first message comprises a device identifier of a third network device and state information; if the state information indicates that the third network device is unreachable from the second network device, the forwarding chip of the first network device looking up a first forwarding entry where the egress interface is a first interface and the device identifier is the device identifier of the third network device; and setting the state of the first forwarding entry to be an invalid state. Therefore, when a fault occurs in a link between network devices, route switching can be carried out in a more timely manner.