Neural strategies to handle routing in computer networks

Area of Science:

• Computer Science
• Artificial Intelligence
• Network Engineering

Background:

• Traditional routing in packet switching networks faces challenges balancing centralized control and distributed adaptability.
• Existing routing strategies often struggle with efficiency and scalability under heavy network loads.

Purpose of the Study:

• To develop a hybrid routing strategy for packet switching networks by integrating neural networks.
• To evaluate the performance and resource requirements of novel neural network-based routing solutions.

Main Methods:

• A neural approach is proposed, incorporating a neural network (N/N) at each node for adjacent routing computations.
• Two distributed routing solutions utilizing an optimizing network and a mapping network are presented.
• Performance is evaluated by comparing neural strategies against classical distributed and centralized routing, considering network overloading effects.

Main Results:

• The neural routing strategies demonstrate competitive performance compared to conventional methods.
• Evaluation includes the impact of additional traffic and resource utilization for each approach.
• The study quantifies the effectiveness of the proposed neural routing in mitigating overloading.

Conclusions:

• Neural network integration offers a viable hybrid approach for packet switching network routing.
• The proposed methods provide a balance between centralized and distributed routing advantages.
• Further research can explore advanced neural architectures for enhanced network performance and efficiency.