Network Simulator for QUIC benchmarking

This project builds a test framework that can be used for benchmarking and measuring the performance of QUIC implementations under various network conditions. It uses the ns-3 network simulator for simulating network conditions and cross-traffic, and for bridging the real world with the simulated world. It uses docker for isolating and coercing traffic between the client and server to flow through the simulated network.

Framework

The framework uses docker-compose to compose three docker images: the network simulator (as found in the sim directory), and a client and a server (as found in the individual QUIC implementation directories, or for a simple shell, the endpoint directory).

The framework uses two networks on the host machine: leftnet (IPv4 193.167.0.0/24, IPv6 fd00:cafe:cafe:0::/64) and rightnet (IPv4 193.167.100.0/24, IPv6 fd00:cafe:cafe💯:/64). leftnet is connected to the client docker image, and rightnet is connected to the server. The ns-3 simulation sits in the middle and forwards packets between leftnet and rightnet.

      +-----------------------+
      |      client eth0      |
      |                       |
      |     193.167.0.100     |
      | fd00:cafe:cafe:0::100 |
      +----------+------------+
                 |
                 |
      +----------+------------+
      |     docker-bridge     |
      |                       |
      |      193.167.0.1      |
      |  fd00:cafe:cafe:0::1  |
+-----------------------------------+
|     |         eth0          |     |
|     |                       |     |
|     |      193.167.0.2      |     |
|     |  fd00:cafe:cafe:0::2  |     |
|     +----------+------------+     |
|                |                  |
|                |                  |
|     +----------+------------+     |
|     |         ns3           |     |
|     +----------+------------+     |
|                |                  |
|                |                  |
|     +----------+------------+     |
|     |         eth1          |     |
|     |                       |     |
|     |     193.167.100.2     |     |
|     | fd00:cafe:cafe:100::2 |  sim|
+-----------------------------------+
      |     docker-bridge     |
      |                       |
      |     193.167.100.1     |
      | fd00:cafe:cafe:100::1 |
      +----------+------------+
                 |
                 |
      +----------+------------+
      |      server eth0      |
      |                       |
      |    193.167.100.100    |
      |fd00:cafe:cafe:100::100|
      +-----------------------+

Building your own QUIC docker image

The endpoint directory contains the base Docker image for an endpoint Docker container. The pre-built image of this container is available on dockerhub.

Follow these steps to set up your own QUIC implementation:

Create a new directory for your implementation (say, my_quic_impl). You will create two files in this directory: Dockerfile and run_endpoint.sh, as described below.

Copy the Dockerfile below and add the commands to build your QUIC implementation.

FROM martenseemann/quic-network-simulator-endpoint:latest

# download and build your QUIC implementation
# [ DO WORK HERE ]

# copy run script and run it
COPY run_endpoint.sh .
RUN chmod +x run_endpoint.sh
ENTRYPOINT [ "./run_endpoint.sh" ]

Now, copy the script below into run_endpoint.sh, and add commands as instructed. Logs should be recorded in /logs for them to be available after simulation completion (more on this later).

#!/bin/bash

# Set up the routing needed for the simulation
/setup.sh

# The following variables are available for use:
# - ROLE contains the role of this execution context, client or server
# - SERVER_PARAMS contains user-supplied command line parameters
# - CLIENT_PARAMS contains user-supplied command line parameters

if [ "$ROLE" == "client" ]; then
    # Wait for the simulator to start up.
    /wait-for-it.sh sim:57832 -s -t 30
    [ INSERT COMMAND TO RUN YOUR QUIC CLIENT ]
elif [ "$ROLE" == "server" ]; then
    [ INSERT COMMAND TO RUN YOUR QUIC SERVER ]
fi

For an example, have a look at the quic-go setup or the quicly setup.

Running a Simulation

From the quic-network-simulator directory, first build the necessary images:
```
CLIENT=[client directory name] \
SERVER=[server directory name] \
docker-compose build
```
Note that you will need to run this build command any time you change the client or server implementation, Dockerfile, or run_endpoint.sh file.

For instance:
```
CLIENT="my_quic_impl" \
SERVER="another_quic_impl" \
docker-compose build
```
You will want to run the setup with a scenario. The scenarios that are currently provided are listed below:
- Simple point-to-point link, with configurable link properties
- Single TCP connection running over a configurable point-to-point link
You can now run the experiment as follows:
```
CLIENT=[client directory name] \
CLIENT_PARAMS=[params to client] \
SERVER=[server directory name] \
SERVER_PARAMS=[params to server] \
SCENARIO=[scenario] \
docker-compose up
```
SERVER_PARAMS and CLIENT_PARAMS may be omitted if the corresponding QUIC implementations do not require them.

For instance, the following command runs a simple point-to-point scenario and specifies a command line parameter for only the client implementation:
```
CLIENT="my_quic_impl" \
CLIENT_PARAMS="-p /10000.txt" \
SERVER="another_quic_impl" \
SCENARIO="simple-p2p --delay=15ms --bandwidth=10Mbps --queue=25" \
docker-compose up
```
A mounted directory is provided for recording logs from the endpoints. docker-compose creates a logs/server and logs/client directory from the directory from which it is run. Inside the docker container, the directory is available as /logs.

Debugging and FAQs

With the server (similarly for the client) up and running, you can get a root shell in the server docker container using the following:
```
docker exec -it server /bin/bash
```

marten-seemann/quic-network-simulator

Network Simulator for QUIC benchmarking

Framework

Building your own QUIC docker image

Running a Simulation

Debugging and FAQs