LSP Transport Hierarchy for Data Center Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data centers face challenges in managing routing and forwarding information due to the massive number of virtual machines and servers, leading to inefficiencies in large-scale data centers, particularly with continuous migration of virtual machines between prefixes or regions.
Innovation Solution
A high-performance, scalable data center switch fabric is introduced, which separates routing and forwarding information between a transport layer and a service layer, using Border Gateway Protocol (BGP) labeled unicast to construct a hierarchy of label switched paths (LSPs) within the transport layer, allowing routers to allocate and modify labels and next hops based on region attributes, thereby reducing the need for extensive label information distribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional L3 aggregation techniques and IP prefix routing are used throughout the data center, then routing information can be maintained in a standardized manner, but the forwarding information bases become excessively large and difficult to manage in large-scale data centers with thousands or millions of virtual machines
Solution Approach 1:
The patent segments the data center network into multiple regions (e.g., leaf-spine regions, rack regions) and creates separate LSP hierarchies for each region. Instead of maintaining a single global routing table with all VM prefixes, each router maintains forwarding information only for its local region and uses LSP labels to reach other regions. This segmentation dramatically reduces the FIB size at each device while maintaining the ability to route traffic across the entire data center.
Solution Approach 2:
The patent introduces region border routers as intermediary devices that aggregate routing information from multiple regions and maintain LSP paths to remote regions. These border routers act as mediators between local leaf routers and remote destinations, allowing leaf routers to forward traffic to border routers which then handle the long-distance routing. This intermediary layer eliminates the need for every router to maintain complete end-to-end routing information.
2Ease of operation
If a flat L2 network is used to connect all virtual machines, then simplified forwarding is achieved, but the network cannot support virtual machine migration between different prefixes or regions
Solution Approach 1:
The patent implements dynamic LSP path establishment and modification based on VM migration events. When a VM migrates between regions or prefixes, the control plane dynamically updates the LSP routing paths to reflect the new location. The BGP-LU protocol automatically detects prefix changes and redistributes routing information to affected routers, allowing the network to adapt to VM mobility while maintaining efficient labeled forwarding.
Solution Approach 2:
The patent changes the forwarding paradigm from L2 MAC address-based forwarding to L3 IP prefix-based forwarding with MPLS labels. This parameter change allows the network to maintain simple label-switched forwarding paths while supporting VM migration through BGP route advertisement updates. The MPLS label stack enables hierarchical routing where inner labels represent local VM prefixes and outer labels represent regional LSP paths.
3Loss of information
If extensive BGP-LU route advertisements are distributed to all routers throughout the data center, then complete routing information is available everywhere, but the processing overhead and memory requirements become unsustainable in large-scale deployments
Solution Approach 1:
The patent applies local quality by having each router maintain complete and accurate routing information for its local region while obtaining routing information for remote regions through LSP label mappings rather than full route advertisements. Leaf routers only need to know how to reach their local VMs and which border router to use for remote traffic; they do not need complete BGP routing tables for all regions. This localized information storage reduces memory requirements while maintaining routing completeness through the hierarchical LSP structure.
4Measurement precision
If the transport layer maintains detailed forwarding information for all service layer endpoints, then precise routing is achieved, but the system becomes difficult to scale to millions of virtual machines
Solution Approach 1:
The patent adds a hierarchical dimension to the routing architecture by introducing multiple levels of LSP nesting. Instead of a flat routing table that scales linearly with the number of VMs, the patent creates a tree-like hierarchy where LSP paths are nested within regions which are nested within the global data center network. This dimensional change allows routing precision to be maintained at each hierarchical level while the overall system scales logarithmically rather than linearly with the number of endpoints.
Data Source
AI summary
A high-performance, scalable data center switch fabric and infrastructure is described that provides a clean separation between of routing and forwarding information between a transport layer of the data center and a service layer of data center that includes the endpoint devices, such as virtual or physical machines deployed within the data center. For example, techniques are described that enable the automatic creation of a transport hierarchy of label switched paths (LSPs) across the transport layer of the data center. For example, the techniques may be applied with the Border Gateway Protocol (BGP) labeled unicast (LU) in a manner that constructs a hierarchy of LPSs in the transport layer of the data center.