Multi-Level Interconnection Network for Data Center Scalability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional tree-based network architectures in data centers face scalability issues due to bandwidth bottlenecks and are susceptible to single points of failure, which can disconnect thousands of servers from the network.
Innovation Solution
A multi-level interconnection network is introduced, where basic cells are aggregated into higher-level cells, enabling fully connected links between servers and using routing algorithms to route messages efficiently, while also employing a fault-tolerant deployment technique to ensure high connectivity without reconfiguring existing network portions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a tree-based network architecture is used to support server interconnection, then the network can be established with a hierarchical structure, but bandwidth bottlenecks occur at core switches and rack switches
Solution Approach 1:
The network is segmented into multiple fully connected cell layers instead of a single hierarchical tree structure. Each cell contains a subset of servers that are fully interconnected, distributing the bandwidth load across multiple parallel paths and eliminating the single bottleneck at core switches.
Solution Approach 2:
The patent transitions from a two-level hierarchical tree structure to a multi-dimensional fully connected cell structure. This dimensional change creates multiple independent communication paths between servers, allowing traffic to flow through alternative routes when one path is congested, thereby increasing overall bandwidth capacity.
2Reliability
If redundant switches are deployed to reduce single point of failure risk, then fault tolerance is improved, but device complexity and cost increase
Solution Approach 1:
The network is divided into multiple fully connected cells, where each cell operates as an independent fault domain. A failure in one cell does not propagate to other cells, providing inherent fault tolerance without requiring extensive redundancy. The segmentation isolates failures and maintains network connectivity through alternative cells.
Solution Approach 2:
The fully connected cell structure provides built-in fault cushioning by creating multiple redundant paths before failures occur. When a switch or server fails, the pre-established alternative paths within the fully connected topology automatically absorb the failure impact, maintaining network operation without requiring active redundancy management.
3Quantity of substance
If more high-level switches are added to support rapid server growth, then network capacity increases, but cost and complexity increase exponentially
Solution Approach 1:
Each fully connected cell serves multiple functions simultaneously: it provides local server interconnection, acts as a fault domain, and enables horizontal scaling. This universal design allows the network to support server growth by simply adding more cells rather than upgrading to more expensive high-level switches with greater functionality.
Solution Approach 2:
The network architecture is designed to dynamically scale by adding identical fully connected cells rather than upgrading to higher-capacity switches. This dynamic approach allows incremental expansion where each new cell immediately contributes to network capacity without requiring complex reconfiguration or expensive hardware upgrades.
4Adaptability or versatility
If a tree-based architecture is used to achieve scalability, then incremental expansion is possible, but the architecture becomes susceptible to single points of failure
Solution Approach 1:
The network is segmented into multiple fully connected cells that can be independently added or removed. This segmentation allows incremental scalability while each cell maintains fault independence, so that adding capacity does not create new single points of failure.
Solution Approach 2:
Instead of creating a single large hierarchical network with centralized control points, the patent uses multiple partial fully connected cells that collectively provide excessive connectivity. This partial approach in each cell, when combined with others, creates an overall network that is both scalable and resilient to individual failures.
Data Source
AI summary
A method and system for providing a multi-level interconnection network is provided. A multi-level interconnection network comprises basic cells that are aggregated into higher level cells at each level of the network. At the first level, the basic cells are aggregated into first level cells. Each first level cell is an aggregation of a number of basic cells that is one more than the number of devices in a basic cell. The basic cells of a first level cell are fully connected; that is, each basic cell has a first level link or connection to each other basic cell. In a first level cell, each device of a basic cell has a first level link to each other basic cell. The multi-level interconnection network has higher level cells that are aggregations of lower level cells in a similar manner.