Selector List for Fast Child Link Failure Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In network devices, child link failures in aggregate Ethernet or SONET bundles lead to slow convergence and traffic loss due to the need for extensive re-programming of packet forwarding engines, which is exacerbated by the proportional increase in memory reads required for load balancing across multiple links.
Innovation Solution
Implementing a selector list that includes indices of child nexthops associated with failed links, allowing for immediate removal of the failed link index, and creating a distribution table based on probabilities to rearrange packet forwarding, enabling fast convergence and efficient load balancing across available links.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all child nexthops are re-programmed in the PFE data plane when a child link fails, then the network device can maintain routing accuracy, but convergence time increases and traffic loss occurs
Solution Approach 1:
The patent segments the nexthop information into two parts: aggregate nexthop information stored in the routing table and child nexthop information stored separately in the PFE data plane selector list. When a child link fails, only the selector list needs to be updated, not the entire routing table, thus maintaining routing accuracy while reducing convergence time
Solution Approach 2:
The patent extracts child nexthop indices from the main routing table and stores them in a separate selector list structure in the PFE data plane. This extraction allows independent updating of child link status without affecting the main routing table, enabling fast convergence while preserving routing accuracy
2Speed
If a selector list with child nexthop indices is implemented, then convergence speed improves during link failures, but the complexity of the data structure increases
Solution Approach 1:
The selector list structure serves multiple functions: it stores child nexthop indices for fast lookup, enables rapid updates when links fail, and maintains the mapping between aggregate and child nexthops. This multi-functionality justifies the additional data structure while delivering fast convergence
3Productivity
If load balancing distributes packets across multiple child links, then network throughput increases, but the number of memory reads required increases proportionally
Solution Approach 1:
The patent performs preliminary action by pre-computing the distribution table from the selector list before packet forwarding. This allows the forwarding process to simply lookup pre-calculated distribution values rather than computing load balancing decisions for each packet, reducing memory reads while maintaining throughput
Data Source
AI summary
A network device provides a selector list that includes indices of child nexthops associated with the network device, where each of the child nexthops is associated with a corresponding child link provided in an aggregated bundle of child links. The network device also receives an indication of a failure of a child link in the aggregated bundle of child links, and removes, from the selector list, an index of a child nexthop associated with the failed child link. The network device further receives probabilities associated with the child links of the aggregated bundle of child links. Each of the probabilities indicates a probability of a packet exiting the network device on a child link. The network device also creates a distribution table based on the probabilities associated with the child links, and rearranges values provided in the distribution table.