Fabric Network Partitioning for Horizontal Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Scaling fabric networks to support a large number of hosts is challenging due to difficulties in storing information about each host across multiple map-servers/fabric control planes, especially with Layer-2 extensions, which can deplete TCAM resources and constrain the network's size.
Innovation Solution
The fabric network is partitioned into smaller segments, with each border node acting as a map-server for a specific partition, allowing for horizontal scaling and reducing the load on each node by registering and managing only a subset of hosts, enabling seamless connectivity and policy enforcement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If fabric network stores information about each host at multiple map-servers, then host location information is maintained, but the network scalability is limited due to TCAM resource depletion
Solution Approach 1:
The fabric network is divided into multiple partitions, with each partition managed by a specific border node as a map-server. This segmentation allows the network to scale horizontally by adding more partitions and border nodes, rather than requiring all border nodes to store information about all hosts. Each border node only maintains location information for hosts within its assigned partition, reducing TCAM resource requirements while maintaining reliable host location tracking.
2Quantity of substance
If Layer-2 extensions are used to support more hosts, then network capacity increases, but TCAM resources are depleted and network size is constrained
Solution Approach 1:
By segmenting the fabric network into multiple partitions and assigning each partition to a specific border node, the system can support a larger total number of hosts across the network without depleting TCAM resources at any single border node. The TCAM resources are distributed across multiple border nodes, each handling only a subset of hosts, thereby enabling network capacity to scale without proportional TCAM resource depletion at individual nodes.
Solution Approach 2:
The patent introduces a new dimension of organization by dividing the network into multiple partitions with different border nodes acting as map-servers. Instead of a single flat structure where all border nodes must know all hosts, the system creates a hierarchical or distributed structure where each border node is responsible for a specific partition. This dimensional change allows the network to scale to support more hosts without requiring each border node to have proportional TCAM resources.
3Device complexity
If a single border node acts as map-server for all hosts, then simplified management is achieved, but the border node becomes overwhelmed and network size is constrained
Solution Approach 1:
The patent segments the map-server functionality across multiple border nodes, with each border node responsible for a specific partition of hosts. This distribution of responsibilities prevents any single border node from becoming overwhelmed while maintaining manageable complexity through clear partition assignments. The system achieves scalability by adding more partitioned border nodes rather than increasing the burden on a single node.
Data Source
AI summary
A method for establishing a partitioned fabric network is described. The method includes establishing a fabric network including a plurality of border nodes to couple the fabric network to one or more external data networks and a plurality of edge nodes to couple to the fabric network to one or more hosts. The method further includes defining a plurality of partitions of the fabric network. The method further includes registering each of the plurality of partitions with a corresponding one of the plurality of border nodes and with each of the plurality of edge nodes.