Distributed DCN Architecture Eliminates Core Switch Bottleneck
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data center networks (DCNs) based on Clos architecture face bottlenecks due to high port density and capacity requirements of core switches, particularly as east-west traffic increases, necessitating a new architecture to distribute switching capacity effectively.
Innovation Solution
A K-level MESH structure is introduced in the DCN, where each switch is interconnected with K other switches, distributing the network-wide switching capacity and eliminating the role of a core switch, using K interconnection port groups and cyclic arrayed waveguide gratings (CAWGs) for optical connections, allowing for one-hop or multiple-hop reachable paths between switches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If an aggregated Clos architecture is used with core switches, then network connectivity is achieved, but port density and core switch capacity become bottlenecks for network-wide switching capacity
Solution Approach 1:
The patent divides the traditional aggregated Clos architecture into multiple distributed data exchange networks (DENs). Each DEN operates independently with its own switches, eliminating the need for a single high-capacity core switch. This segmentation distributes the switching capacity across multiple smaller switches, thereby avoiding the port density bottleneck at the core switch level while maintaining overall network-wide switching capacity.
2Productivity
If core switch capacity is increased to handle east-west traffic, then network throughput improves, but the bottleneck at core switch capacity becomes more pronounced
Solution Approach 1:
The patent combines multiple distributed data exchange networks into a federated architecture where each DEN contributes its switching capacity to the overall network. Instead of relying on a single core switch to handle all east-west traffic, the system merges the capabilities of multiple smaller switches across different DENs, thereby achieving high network throughput without requiring any single core switch to have excessive capacity.
3Adaptability or versatility
If a distributed MESH structure is implemented, then network scalability improves and core switch bottleneck is eliminated, but network topology complexity increases
Solution Approach 1:
The patent introduces a hierarchical dimension to the distributed MESH structure by organizing switches into multiple levels within each DEN and establishing inter-DEN connections. This dimensional organization provides clear routing paths and simplifies topology management compared to a flat MESH structure, thereby maintaining network scalability while reducing the perceived complexity through structured layering.
Data Source
AI summary
A DCN includes N first-level subnetworks, each first-level subnetwork includes N second-level subnetworks, each kth-level subnetwork includes N (k+1)th-level subnetworks, each (K−1)th-level subnetwork includes multiple switches, and the DCN is a K-level network; each switch in the DCN has K subnetwork identifiers, the K subnetwork identifiers are respectively used to indicate each level of subnetwork to which the switch belongs and a number in a (K−1)th-level subnetwork to which the switch belongs; and the switch is separately interconnected with each switch in K direct connection switch groups, each direct connection switch group includes N−1 switches, and an ith-level subnetwork identifier of the N−1 switches included in an ith direct connection switch group of the K direct connection switch groups is different from an ith-level subnetwork identifier of the switch, and is the same as other K−1 subnetwork identifiers of the switch.