Optical Bypass Switch Neural Network Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers face bottlenecks and increased power consumption due to the need for electrical switches to determine packet destinations on a per-packet basis, which slows down communication, especially when handling large amounts of data.
Innovation Solution
A hybrid network system incorporating both electrical packet switches and optical circuit switches, where neural networks analyze packet flows to identify 'elephant flows' and direct them through high-speed, high-bandwidth optical paths, while electrical switches handle lower-bandwidth traffic, reducing congestion and optimizing data routing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If electrical packet switches handle all traffic, then packet routing flexibility is maintained, but processing speed and power efficiency deteriorate due to per-packet destination determination
Solution Approach 1:
The patent segments traffic into two categories: elephant flows (high bandwidth) and mouse flows (low bandwidth). Optical circuit switches handle elephant flows with high-speed fixed-path routing, while electrical packet switches handle mouse flows with flexible per-packet routing. This segmentation resolves the contradiction by applying different switching mechanisms to different traffic types, achieving high speed for appropriate traffic without requiring all traffic to undergo complex electrical switching.
Solution Approach 2:
The system dynamically identifies and classifies flows using neural networks to distinguish elephant flows from mouse flows. The optical circuit switch configuration is dynamically adjusted based on flow classification results, enabling the system to adaptively route high-bandwidth traffic through high-speed optical paths while maintaining electrical packet switching for other traffic. This dynamic adaptation resolves the speed-complexity contradiction.
2Productivity
If optical circuit switches are used for high-bandwidth traffic, then data transmission speed improves, but neural network analysis and flow classification complexity increases
Solution Approach 1:
The patent replaces traditional mechanical flow classification methods (such as hardware flow counters) with neural network-based classification. The neural network analyzes packet flows and determines whether they are elephant or mouse flows, enabling more accurate and adaptive classification. This substitution increases productivity by correctly identifying high-bandwidth traffic for optical routing while managing classification complexity through the efficiency of neural network processing.
3Use of energy by moving object
If electrical packet switches process all packets, then routing flexibility is maintained, but power consumption increases due to continuous per-packet processing
Solution Approach 1:
The patent segments traffic handling between two types of switches: optical circuit switches for elephant flows and electrical packet switches for mouse flows. By routing high-bandwidth elephant flows through optical switches, the system dramatically reduces power consumption since optical switches do not require per-packet processing. Meanwhile, electrical packet switches continue to provide routing adaptability for mouse flows. This segmentation resolves the energy-routing flexibility contradiction.
Solution Approach 2:
The patent extracts elephant flow traffic from the electrical packet switching domain and routes it through optical circuit switches. This extraction removes the power-intensive per-packet processing requirement from high-bandwidth traffic while preserving routing flexibility for remaining traffic through the electrical packet switches. The neural network enables this extraction by identifying which flows should be separated from the electrical switching path.
Data Source
AI summary
A flow of packets is communicated through a data center including an electrical switch, an optical switch, and multiple racks each including multiple network devices. The optical switch can be controlled to receive packet traffic from a network device via a first optical link and to output that packet traffic to another network device via a second optical link. One network device includes a neural network that analyzes received packets of the flow. The optical switch is controlled to switch based on a result of the analysis performed. In one instance, the optical switch is controlled such that immediately prior to the switching no packet traffic passes from the first optical link and through the optical switch and to the second optical link but such that after the switching packet traffic does pass from the first optical link and through the optical switch and to the second optical link.