Discretized learning automata solutions to the capacity assignment problem for prioritized networks

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• Traditional network capacity assignment (CA) often simplifies traffic to a single class.
• Real-world networks transmit diverse packet classes with varying lengths and priorities.
• The CA problem is computationally challenging, classified as NP-hard.

Purpose of the Study:

• To introduce a novel discretized learning automaton (LA) approach for network capacity assignment.
• To address the generalized CA problem with multiple prioritized traffic classes.
• To minimize network costs while meeting all traffic demands.

Main Methods:

• Implementation of a discretized learning automaton (LA) solution.
• Introduction of a new 'meta-action' philosophy within the LA framework.
• Comparison against existing heuristic, simulated annealing, and genetic algorithm (GA) methods.

Main Results:

• The proposed discretized LA solution effectively handles multi-class, prioritized network traffic.
• The method demonstrates superior performance compared to previous continuous LA and heuristic approaches.
• The novel meta-action philosophy enhances the LA's problem-solving capabilities.

Conclusions:

• The discretized LA offers a highly effective solution for the complex network capacity assignment problem.
• This approach provides a significant advancement for optimizing network resources with diverse traffic.
• The meta-action philosophy represents a key innovation in learning automaton applications.