Dynamic neural-based buffer management for Queuing systems with self-similar characteristics.

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• Buffer management in queuing systems is crucial for balancing network efficiency (packet loss) and fairness (individual source packet loss).
• Complete partitioning (CP) and complete sharing (CS) represent extreme approaches, while dynamic partitioning seeks a compromise.
• Existing dynamic techniques face challenges in implementation practicality and handling complex traffic patterns like self-similarity.

Purpose of the Study:

• To introduce two novel dynamic buffer management techniques for queuing systems.
• To address the tradeoff between network efficiency and fairness, particularly for self-similar traffic.
• To leverage perceptron neural networks for adaptive traffic pattern learning and water-filling approaches.

Main Methods:

• Development of two new dynamic buffer management techniques.
• Application of perceptron neural networks to learn and adapt to arriving traffic patterns.
• Utilization of the water-filling approach to manage buffer allocation.
• Extensive computer simulations to evaluate performance.

Main Results:

• The proposed techniques demonstrate effective management of the efficiency-fairness tradeoff.
• Both techniques show excellent efficiency and fairness characteristics in simulations.
• The methods are shown to be easy to implement in practical queuing systems.
• Successful accommodation of self-similar traffic patterns was achieved.

Conclusions:

• The novel dynamic buffer management techniques offer a practical solution for optimizing queuing systems.
• Perceptron neural networks provide adaptive learning capabilities beneficial for traffic management.
• The water-filling approach combined with neural networks effectively balances efficiency and fairness.
• These techniques present a promising advancement in managing complex network traffic.