Spine Node Partial Replication for Network Scalability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional leaf-spine hierarchical networks face scalability and performance challenges due to the need for each spine node to maintain full routing information for all endpoints, leading to hotspot issues and inefficient resource utilization.
Innovation Solution
A partial replication system is implemented, where endpoint routing information is divided and distributed across multiple spine nodes using a sharding protocol, allowing each spine node to maintain a subset of routing information based on its capacity and load, with shard mapping information propagated to leaf nodes to ensure efficient traffic forwarding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If each spine node maintains full routing information for all endpoints, then routing completeness is improved, but network bandwidth and CPU loads increase
Solution Approach 1:
The patent divides the routing information into segments distributed across multiple spine nodes using a sharding protocol. Each spine node maintains routing information for a specific subset of endpoints (shard) rather than all endpoints, thereby reducing the routing table size and associated CPU/memory loads while maintaining overall routing completeness through the collective capacity of the spine layer.
2Reliability
If each spine node maintains full routing information for all endpoints, then routing completeness is improved, but scalability deteriorates
Solution Approach 1:
By segmenting routing information across multiple spine nodes, the system enables linear scalability. When network capacity needs to increase, additional spine nodes can be added to handle more endpoint shards without requiring existing nodes to process the full routing table, thus maintaining routing completeness while improving scalability.
Solution Approach 2:
The patent introduces a new dimension to routing information management by implementing a sharding protocol that distributes routing data across the spine layer in a distributed manner. This dimensional change from centralized full replication to distributed segmentation enables the network to scale horizontally by adding more spine nodes, each handling a portion of the total routing information.
3Adaptability or versatility
If routing information is distributed across multiple spine nodes using sharding, then scalability is improved, but device complexity increases
Solution Approach 1:
The sharding protocol implementation is designed to be universal across all spine nodes in the network. Each spine node runs the same sharding logic and maintains routing information for its assigned endpoints using standardized data structures and protocols. This multi-functionality approach, where each node performs similar operations on its subset of data, reduces the complexity burden compared to asymmetric or highly customized solutions.
Data Source
AI summary
Embodiments are provided for partially replicating endpoint routing information, and comprise calculating a first shard interval of a key space based, at least in part, on capacities of a plurality of spine nodes in a network fabric. Embodiments also include mapping the first shard interval to a first spine node of the plurality of spine nodes, communicating shard mapping information associated with the mapping to a set of leaf nodes in the network fabric, and populating an endpoint repository in the first spine node with routing information for one or more endpoints corresponding to the first shard interval. More specific embodiments include calculating respective shard intervals for other spine nodes of the plurality of spine nodes based, at least in part, on the capacities of the plurality of spine nodes. In specific embodiments, the calculating the first shard interval is based, in part, on one or more dynamic parameters.