🧠 LLM-Powered Network Slice Negotiation Simulator

This repository contains an asynchronous Python simulation framework that models multi-agent negotiation for resource allocation in 5G/6G network slicing.

It simulates a negotiation process between two autonomous agents representing:

• 📶 eMBB (Enhanced Mobile Broadband)
• ⚡ URLLC (Ultra-Reliable Low-Latency Communication)

These agents leverage a local Large Language Model (LLM) to:

• Reason about network conditions
• Propose bandwidth allocations
• Generate counter-proposals

All decisions are guided by a Digital Twin predictor, with the goal of:

✅ Minimizing energy consumption
✅ Strictly adhering to Service Level Agreements (SLAs)

---

🚀 Features

🤖 LLM-Driven Autonomous Agents

• Uses a local LLM for reasoning and negotiation
• Dynamic proposal generation with concession pressure modeling

⚡ 1B Model Optimization (otel-llm-1b-it)

• Tailored for small, fast local models
• Handles quirks of low-parameter LLMs

📊 Predictive Digital Twin (DT)

• M/M/1 queueing model
• CVaR (Conditional Value at Risk) for tail latency prediction
• Estimates:
  • Latency
  • Energy consumption

🔄 Asynchronous Negotiation Loop

• Built with asyncio
• Enables parallel LLM calls
• Reduces simulation runtime significantly

🛡️ Robust Fallback Mechanisms

• Regex-based parsing for malformed outputs
• Handles:
  • JSON failures
  • Schema hallucinations
  • String-to-number conversion issues

📈 Proportional Climb/Descent Logic

• Prevents:
  • Deadlocks
  • Overshooting
• Uses PID-like adjustments based on SLA violations

📉 Rich Metrics & Visualization

• Tracks:
  • Latency
  • SLA violations
  • Utility degradation
  • Energy savings
• Generates CDF plots using matplotlib

---

🛠️ Prerequisites & Installation

1. Environment Setup

Ensure you have Python 3.8+ installed.

Install dependencies:

pip install numpy matplotlib openai