Bandwidth Demand Estimator (BDE)

The Bandwidth Demand Estimator (BDE) is a network function designed to efficiently manage bandwidth allocation in a Network Slice (NS) to meet packet delay requirements while minimizing Quality of Service (QoS) violations. This repository contains the implementation of the BDE along with tools for scenario generation using the Amarisoft testbed.

Overview

The BDE utilizes a data-driven approach, incorporating QoS feedback to dynamically adjust bandwidth allocation at the base station (AMARI Callbox). It employs reinforcement learning techniques to learn the optimal bandwidth allocation policy while considering the future impact of its actions on packet queues and QoS metrics.

Key Features

• Reinforcement Learning Approach: The BDE employs reinforcement learning, where actions represent allocated bandwidth, states describe traffic conditions, wireless channel status, and packet queue information, and the cost function is a weighted sum of bandwidth allocation and QoS violations.

• Transition Matrix Estimation: Periodic estimation of the transition matrix facilitates value iteration to find the optimal policy. Multi-armed bandits are utilized for initial estimation, leveraging the monotonicity of the cost function with respect to actions.

• Integration with Amarisoft Testbed: The implementation interfaces with the Amarisoft API, providing tools for scenario generation and performance evaluation.

Components

1. BDE Implementation (main.py)

• Algorithms for bandwidth allocation based on reinforcement learning.
• Transition matrix estimation and value iteration.
• Integration with QoS feedback mechanisms.

2. Scenario Generation Tool (amarify.py)

• Generates test scenarios using the Amarisoft testbed.
• Configures parameters for LTE cell settings, traffic types, and UE behaviors.
• Utilizes JavaScript Object Notation (JSON) configuration files for scenario definition.

Amarisoft Testbed Integration

The Amarisoft testbed consists of the AMARI Callbox Ultimate and the AMARI UE Simbox, facilitating over-the-air LTE communication with SDRs. The BDE implementation runs on an Ubuntu PC connected to the testbed, enabling configuration, traffic monitoring, and scenario execution via SSH commands.

Prerequisites

1. Install Python 3 and Node.js (tested with Python 3.10.12 and Node.js v20.5.0)
2. Install python dependencies: pip install matplotlib numpy paramiko.
3. Install the WebSocket client and server implementation for Node.js: npm install nodejs-websocket.

Usage

To utilize the BDE and generate test scenarios: 1. Configure scenarios using python3 amarify.py. This will produce the following: - Configuration file ue_db-test.cfg to be placed under ~/mme/config in the AMARI Callbox node. - Configuration file ue-test.cfg to be placed under ~/ue/config in the AMARI UE Simbox node. - A plot user_activity.pdf visualizing the number of active users per time unit. 2. Execute the BDE implementation python3 main.py on the computing node (e.g., Ubuntu PC) connected to the Amarisoft testbed, which includes the main simulation loop. - This produces the bandwidth estimate for the network slice, which is updated using the Amarisoft API through util/liveness.py. - Valid PRB values are of the form 2^n₁ · 3^n₂ · 5^n₃, where n₁, n₂, n₃ ∈ ℕ. - Simulation parameters including idle time between selecting arm and estimating reward are set in util/parameters.py. - IMPORTANT: Networking parameters for accessing the remote log files from the two Amarisoft nodes are defined in remote_params() in util/logflatten.py. The nodes are accessed in non-interactive fashion using SSH key-based authentication. 3. To optionally run a script as part of the util package, use e.g., python3 -m util.liveness.

Citation

If you use this implementation, please cite the following paper: [P. Nikolaidis, A. Zoulkarni, J. Baras, "Data-driven Bandwidth Adaptation for Radio Access Network Slices," arXiv:2311.17347, 2023]

Contributors

• Panagiotis Nikolaidis (nikolaid@umd.edu)
• Asim Zoulkarni (asimz@umd.edu)