butterknife

A network-slicing system for software-defined mobile networks

butterknife is a proof-of-concept system that can prioritize particular user connections over software-defined mobile networks. Using a modified scheduler integrated into a fork of srsRAN, we simulate a full end-to-end LTE connection, including both a software-defined base-station, and software-defined user equipment.

This project was developed as the final project for ECE 257A at University of California, San Diego.

File structure

• butter/ - A network usage simulator, simulating various dynamic video streaming applications. Also provides a notion of quality of experience of the various applications through a Python interface.
• handle/ - Provides a fork of srsRAN, including a modified scheduler that can take external priorities for individual users, and Python interfaces for modifying scheduler behavior.
• slicer/ - A reinforcement-learning based application for intelligent prioritization of users based on application information.
• toast/ - A virtual tunnel interface, used for simulating srsRAN behaviors. Currently not in use.

Usage

The following section details how to run butterknife. First, butterknife uses poetry for dependency management. Please utilize an up-to-date version of poetry.

Each individual folder within this repository is set up as a separate poetry project. Additionally, the srsRAN and srsUE fork in handle needs to be built and started. Please refer to the srsRAN documentation for details on that process. To run the full butterknife system, each project needs to be run separately. The order in which to run the projects are:

1. Start the srsRAN fork and the srsUE.

2. Start butter server

3. Start butter client with the appropriate settings.

4. Start slicer.

5. Start the python interface in handle.

Application notes

The following section provides an overview on how to accomplish the tasks listed above. Note that only a single butter client needs to be started. Additionally, the slicer simulator need only be run when desired.

To run the server:

poetry run python ./butter/butter/server.py

To run the unbuffer client:

poetry run python ./butter/butter/client.py --ip <ip address> --ft <frame time> --rt <recovery time> --bs <buffer size> --unbuffer --id <id>

To run the buffer client:

poetry run python ./butter/butter/client.py --ip <ip address> --ft <frame time> --rt <recovery time> --bs <buffer size> --buffer --id <id>

Default value of parameter for client:

--ip: "local host"

--ft: "0.016"

--rt: "0.016"

--bs: "500"

--buffer/--unbuffer

--id: "70"

UE configurations:

	Buffer	Unbuffer
High throughput	Buffer size: 500, frame time = 0.008	Buffer size: 1, frame time = 0.008
Low throughput	Buffer size: 500, frame time = 0.032	Buffer size: 1, frame time = 0.032

To run the network slicing application (reinforcement learning model):

poetry run python ./slicer/slicer/network_slicing_application.py

To run the network slicing model simulation (reinforcement learning model):

poetry run python ./slicer/slicer/network_slicing_application_sim.py

To run the scheduler interface:

poetry run python ./handle/handle/rl_interface.py

Contact us!

If you have any questions or concerns about this project, feel free to contact us!