Border Leaf Traffic Convergence in Software Defined Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In software defined networking, especially in high-scale data centers, layer 3 conversational learning at border leaf nodes introduces additional latency and delays during failures or topology changes due to the need to install host routes in hardware forwarding tables, which is exacerbated by the use of multiple border leaf nodes for redundancy or load sharing.
Innovation Solution
The technique synchronizes host routes across border leaf nodes, ensuring that active conversations are learned and forwarded without additional latency by pre-installing routes in hardware forwarding tables of other nodes that already host the tenant, using synchronization messages and state distribution within the control plane, thereby minimizing convergence time and delays.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If layer 3 conversational learning is implemented at border leaf nodes, then routing flexibility and adaptability are improved, but convergence time increases significantly during failures or topology changes
Solution Approach 1:
The patent pre-installs host routes in the hardware forwarding tables of border leaf nodes before they are actually needed. When a border leaf node detects an active conversation, it proactively pushes the corresponding host route to its hardware forwarding table and simultaneously notifies other border leaf nodes via BGP synchronization messages, so that backup nodes already have the routes ready in their hardware tables before any failure occurs. This preliminary action eliminates the need for slow route installation during failures.
Solution Approach 2:
The patent prepares backup routing paths in advance by synchronizing host route states across multiple border leaf nodes. When a border leaf node fails, the backup nodes already have the necessary host routes installed in their hardware forwarding tables, providing an immediate cushion against the failure. This is achieved through state synchronization mechanisms that maintain route information across the border leaf node cluster before any actual failure occurs.
2Reliability
If multiple border leaf nodes are deployed for redundancy and load sharing, then system reliability is improved, but convergence time is worsened due to the need to synchronize and install routes across multiple nodes
Solution Approach 1:
The patent implements preliminary route installation in hardware forwarding tables of all border leaf nodes before failures occur. When the control plane determines that a host route is active, it proactively installs the route in the hardware forwarding table of the primary border leaf node and synchronizes the route state to all backup border leaf nodes via BGP. This ensures that all nodes have routes pre-configured in their hardware tables, eliminating the need for time-consuming route installation during failover events.
Solution Approach 2:
The patent uses state copying mechanisms to replicate host route information across multiple border leaf nodes. When a border leaf node learns about an active conversation, it copies the host route state to other border leaf nodes through BGP synchronization messages. This copying approach allows backup nodes to have identical route information in their hardware forwarding tables, enabling immediate failover without route installation delays.
3Productivity
If host routes are installed in hardware forwarding tables during active conversations, then forwarding performance is improved, but additional latency is introduced during failures due to the installation process
Solution Approach 1:
The patent performs preliminary installation of host routes in hardware forwarding tables before failures occur. When the control plane detects an active conversation, it immediately pushes the host route to the hardware forwarding table of the primary border leaf node and synchronizes the route state to all backup nodes. This preliminary action ensures that routes are already in hardware tables with optimized forwarding paths established before any failure occurs, eliminating installation latency during failover.
Solution Approach 2:
The patent merges the route installation process with the normal route advertisement process. Instead of installing routes separately during failures, the control plane combines route synchronization with BGP updates, so that host routes are installed in hardware forwarding tables as part of the normal route advertisement flow. This merging eliminates additional latency by making route installation an integrated part of the routing protocol operation rather than a separate post-failure action.
Data Source
AI summary
A border leaf of a software defined computer network receives network traffic directed to a first endpoint from a second endpoint. The first endpoint is connected to the software defined network, and the second endpoint is external to the software defined network. If the border leaf determines that it does not have a stored network route for an active conversation between the first endpoint and the second endpoint, the border leaf determines a valid network route for the network traffic to reach the first endpoint. The border leaf stores the valid network route and forwards the network traffic to the first endpoint. The border leaf transmits a synchronization message to other border leaf nodes of the computer network. The synchronization message causes the other border leaf(s) to store the valid network route and an indication that the original border leaf is forwarding the network traffic in the active conversation.