Spanning Tree Load Balancing for Data Center Scalability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current load balancing techniques, such as mesh networks, are resource-intensive and impose scaling limits due to the need for all load balancing devices to exchange metrics with each other, leading to inefficient network resource usage and limited scalability.
Innovation Solution
Implementing a spanning tree structure for load balancing devices to exchange metrics, where only necessary connections are maintained, reducing the number of transmissions and conserving network resources by designating an authoritative load balancing device to manage metric exchanges within and between regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a mesh network topology is used for load balancing devices to exchange metrics, then all devices can exchange metrics with each other for comprehensive load balancing information, but network resource consumption increases and scalability is limited due to the O(n²) number of connections required
Solution Approach 1:
The mesh network is segmented into hierarchical regions with regional load balancers managing local segments and a global load balancer coordinating between regions. This segmentation reduces the number of direct connections each device must maintain while preserving comprehensive metrics exchange capability through the hierarchical structure.
Solution Approach 2:
Regional load balancers act as intermediaries between local load balancing devices and the global load balancer. They aggregate metrics from their region and forward to the global load balancer, eliminating the need for direct connections between all devices while maintaining comprehensive information exchange.
2Loss of information
If all load balancing devices exchange metrics with each other in a mesh network, then complete visibility of system state is achieved, but network resource consumption and processing overhead increase significantly
Solution Approach 1:
The patent extracts the metric aggregation function from individual load balancing devices and concentrates it at regional and global load balancer levels. This extraction reduces the number of metric transmissions across the network while maintaining complete system state visibility at the hierarchical aggregation points.
Solution Approach 2:
Metrics from multiple local load balancing devices are merged and aggregated by regional load balancers, which then merge regional metrics at the global load balancer. This merging process reduces redundant information transmission while preserving complete system state information.
3Ease of operation
If a mesh network topology is implemented for load balancing devices, then direct peer-to-peer communication is enabled, but the system reaches scaling limits due to the quadratic growth of connections
Solution Approach 1:
The patent transitions from a flat two-dimensional mesh topology to a three-dimensional hierarchical topology with multiple levels (local devices, regional load balancers, global load balancer). This dimensional change enables the system to scale linearly rather than quadratically while maintaining peer-to-peer communication capabilities within each hierarchical level.
Data Source
AI summary
A device may receive first data center information associated with first data centers in a first region. The device may receive second data center information associated with second data centers in a second region that is different from the first region. The device may generate a spanning tree based on the first data center information and the second data center information. The spanning tree may be associated with exchanging metrics associated with the first data centers and the second data centers, and may be used to exchange the metrics between the device and other devices included in the first data centers or the second data centers. The device may exchange the metrics with the other devices based on the spanning tree.