Data Center Routing Tables With Off-Shortest Path Forwarding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Highly-structured data center networks suffer from increased latency due to multiple hops, underutilization of resources, difficulty in horizontal scaling, and inefficiencies in routing protocols, particularly in source-oriented Link-State and destination-oriented Distance-Vector protocols, leading to high storage and computation overheads.
Innovation Solution
A method for building routing tables in a data center environment using a flat, pseudo-random layer of Top of Rack switches, eliminating intermediate layers, and employing a routing protocol that includes both Shortest Path and Off-Shortest Path routes to ensure loop-free packet forwarding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If highly-structured network architectures (Clos, FatTree, Dragonfly) are used to provide high bisectional bandwidth and resilience, then network reliability and bandwidth are improved, but latency increases due to multiple hops through intermediate layers
Solution Approach 1:
The patent extracts and eliminates the intermediate aggregation and spine layers from the traditional Clos architecture, creating a flat network topology where leaf switches directly interconnect. This removal of intermediate components reduces the number of hops and latency while maintaining network reliability through direct peer-to-peer paths between leaf switches.
Solution Approach 2:
The patent transitions from a hierarchical three-layer structure to a flat single-layer structure, fundamentally changing the dimensional organization of the network. This dimensional change eliminates vertical hops through intermediate layers while maintaining horizontal connectivity, thereby reducing latency without sacrificing reliability.
2Ease of operation
If traditional routing protocols (Link-State or Distance-Vector) are implemented in flat networks, then routing functionality is provided, but storage and computation overheads increase significantly
Solution Approach 1:
The patent extracts the complex routing protocol computations from individual switches and centralizes them in a controller. This extraction eliminates the need for each switch to maintain full routing tables and run complex Link-State or Distance-Vector algorithms, significantly reducing storage and computation overhead at the switch level while maintaining complete routing functionality through controller-based management.
Solution Approach 2:
The patent introduces a controller as an intermediary between switches and the routing protocol. The controller acts as a mediator that computes routes centrally and distributes forwarding instructions to switches, eliminating the need for switches to independently run complex routing protocols. This intermediary approach maintains routing functionality while dramatically reducing storage and computation requirements at the network edge.
3Reliability
If intermediate aggregation and spine layers are added to provide multiple paths for ECMP routing, then network resilience and path diversity are improved, but device complexity and hardware requirements increase
Solution Approach 1:
The patent removes intermediate aggregation and spine layers from the network architecture, eliminating the complexity associated with multi-layer hierarchies. Network resilience is maintained through direct interconnections between leaf switches, which provide multiple paths without requiring intermediate layers. This extraction simplifies the overall device complexity while preserving essential resilience characteristics.
Solution Approach 2:
The patent merges the functions of leaf, aggregation, and spine layers into a single flat layer of directly interconnected switches. This consolidation eliminates the need for separate intermediate layers and reduces architectural complexity while maintaining path diversity and resilience through direct peer-to-peer connections between all leaf switches.
4Ease of operation
If conservative routing (Shortest Path only) is used to ensure loop-free forwarding, then routing simplicity is maintained, but adaptability to network congestion decreases
Solution Approach 1:
The patent implements dynamic routing that adapts to network conditions by allowing packets to deviate from shortest paths when congestion is detected. The system maintains routing simplicity through centralized control while gaining adaptability to congestion through dynamic path selection. Routes are adjusted in real-time based on network state, enabling the system to balance simplicity and adaptability.
Solution Approach 2:
The patent changes the routing parameter from fixed shortest-path metrics to dynamic metrics that incorporate congestion information. The controller adjusts route selection based on real-time network conditions, allowing the system to maintain simplicity through centralized management while adapting to congestion through parameter changes in route selection criteria.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
Systems and methods for building routing tables for use in a data center network are provided. In one embodiment, a switch is arranged in a network environment having a plurality of switches. The switch includes a plurality of inward-directed ports configured for direct connection with a plurality of servers and a plurality of outward-directed ports configured for direct connection with a subset of the plurality of switches. The switch also includes a processor configured to generate a Routing Table (RT) in an RT build process and to create a Forwarding Table (FT) for forwarding packets when the RT build process is complete. The outward-directed ports are configured to forward packets to one or more switches of the subset of switches according to the RT. The FT includes at least a Shortest Path (SP) route and one or more Off-Shortest Path (OSP) routes for forwarding the packets to the one or more switches.