Distributed Control for Photonic Switched Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Photonic networks face challenges in achieving low delay, loss-less switching, and efficient bandwidth allocation due to propagation delays and computational complexity in large-scale systems, particularly in high-speed data centers with multiple server racks and photonic switches.
Innovation Solution
A distributed control architecture with group controllers that monitor and allocate bandwidth across sub-networks, reducing propagation delays and computational complexity by distributing processing and communication across a network of aggregation switch nodes and switch controllers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If centralized control architecture is used for bandwidth allocation in photonic networks, then control decisions can be made quickly, but propagation delay between multiple photonic switches increases and control processing power requirements increase
Solution Approach 1:
The control architecture is segmented into hierarchical levels: core controllers manage inter-group bandwidth allocation while aggregation switch nodes manage intra-group allocation. This segmentation distributes control functions, reducing propagation delay for local decisions while maintaining coordinated global bandwidth management through the hierarchical structure.
Solution Approach 2:
The control architecture introduces a hierarchical dimension with two levels (core controllers and aggregation switch nodes) to manage bandwidth allocation. This dimensional organization allows simultaneous local的快速 response and global coordination, resolving the contradiction between fast control decisions and reduced propagation delay.
2Speed
If distributed control architecture is used to reduce propagation delay, then control processing power requirements increase and coordination between controllers becomes more complex
Solution Approach 1:
Control processing is segmented between core controllers and aggregation switch nodes. Aggregation nodes handle local bandwidth allocation and demand aggregation, reducing the processing burden on core controllers. This segmentation enables faster distributed control while distributing computational power requirements across multiple components.
Solution Approach 2:
Aggregation switch nodes perform partial control functions locally (intra-group bandwidth allocation) rather than requiring all control decisions to be made by core controllers. This partial action at the edge enables faster local responses while reducing overall control processing power requirements through distributed computation.
3Productivity
If bandwidth allocation is performed at submicro-second time frame, then control processing power requirements increase significantly, but network performance improves
Solution Approach 1:
Bandwidth allocation is segmented into hierarchical levels with different time scales. Core controllers operate at longer time scales for inter-group allocation, while aggregation switch nodes handle faster intra-group allocation. This segmentation enables submicro-second response times for local decisions without requiring all controllers to operate at the highest speed, thus reducing overall processing power requirements.
Solution Approach 2:
The system performs partial bandwidth allocation at the aggregation node level with faster response times, while core controllers handle less frequent inter-group allocations. This partial action at multiple levels achieves high network throughput through fast local responses without requiring excessive control processing power across the entire system.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present application provides methods of determining bandwidth demands and allocating bandwidth for a network including a plurality of groups of sub-networks. Each subnetwork includes a plurality of server racks and an aggregation switch node configured to control the flow of traffic to and from the plurality of server racks in the subnetwork. The aggregation switch nodes in the network are optically coupled to one another via at least one switch controlled by a switch controller. Each group of subnetworks of the plurality of groups of subnetworks also includes a group controller in communication with each of the aggregation switch nodes in the group. The group controllers are communicatively coupled to one another. The interaction between the aggregation switch nodes, the group controllers and the switch controllers enable distributed and dynamic control of switching between the subnetworks.