Graceful Restart Master-Slave Segmentation for BGP Deadlock Avoidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In network devices using the Border Gateway Protocol (BGP), a deadlock situation can occur during the Graceful Restart (GR) process when multiple neighboring devices are restarted, leading to traffic interruption due to waiting for routes from each other.
Innovation Solution
The method involves designating a master and slave GR Restarter from multiple GR Restarters, allowing the master to perform route selection without waiting for routes from its slave, and using distinct EOR messages to indicate complete or incomplete route reception, thereby avoiding deadlocks and ensuring uninterrupted traffic.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple neighboring BGP devices are restarted simultaneously using conventional GR process, then each device waits for routes from all neighbors to complete route selection, but this causes a deadlock situation where no device can complete route selection, leading to traffic interruption
Solution Approach 1:
The patent segments the GR process by introducing master-slave relationships among multiple restarting devices. Each device is designated as either a master or slave, and slave devices perform route selection before receiving complete routes from all neighbors. This segmentation breaks the deadlock by allowing devices to proceed in stages rather than waiting for all neighbors simultaneously.
Solution Approach 2:
The patent applies preliminary action by allowing slave devices to perform route selection before receiving complete route information from all neighboring devices. The master device sends a notification to slave devices indicating that route information is available, enabling slaves to proceed with route selection in advance rather than waiting passively for all routes to arrive.
2Reliability
If conventional GR process is used with multiple restarting devices, then each device must wait for End-of-Rib (EOR) messages from all neighbors before completing route selection, but this creates a circular dependency that results in deadlock
Solution Approach 1:
The patent divides the GR process into master and slave roles, where slave devices can complete route selection independently of receiving EOR messages from all neighbors. This segmentation eliminates the circular dependency by creating asymmetric requirements: masters wait for EOR messages, while slaves can proceed based on notifications from masters.
Solution Approach 2:
The patent introduces feedback mechanisms where master devices send notifications to slave devices indicating that route information is available. This feedback enables slave devices to proceed with route selection without waiting for traditional EOR messages from all neighbors, breaking the circular dependency while maintaining route selection completeness.
Data Source
AI summary
This disclosure provides methods and devices for Graceful Restart (GR) in a network with multiple GR Restarters that include a first GR Restarter and at least one second GR Restarter corresponding to the first GR Restarter. After receiving routes from all normal neighbor network devices of the first GR Restarter, the first GR Restarter performs route selection to determine first selected routes, and sends the first selected routes to the at least one second GR Restarter. After receiving routes from the at least one second GR Restarter, the first GR Restarter performs route selection to determine second selected routes, updating a local Forwarding Information Base (FIB) at the first GR Restarter with the second selected routes, and sends the second selected routes to all neighbor network devices of the first GR Restarter. The first GR Restarter directly connects with the at least one second GR Restarter corresponding to the first GR Restarter.