This repository contains the source code of Flexicast QUIC, accepted at SIGCOMM CCR for the April 2025 volume.
Disclamer: we are currently working on an updated version of Flexicast QUIC to contain the last updates from Multipath QUIC and Cloudflare quiche. It will be released soon.
This project builds upon quiche and multipath. The core implementation of flexicast-quic relies on quiche and its multipath branch (see the paper).
The core implementation, quiche, is extended to support flexicast QUIC. The applications are in the apps repository. For this paper, we implemented and used the fc-client-rtp and fc-server-rtp-tokio applications. The other applications are initially part of quiche or rely on applications not evaluated.
Please use the refactoring branch of this project, as other branches contain outdated code. We will make sure to clean as well as possible the repository in the future.
The main contributions inside the code repository of quiche lies in the multicast module. This module contains as much as possible of the control of flexicast to avoid modifying to much the main part of quiche. However, we still had to modify the main functions (in lib.rs) at some point; as well as the frames, the recovery, etc...
The message passing implementation to enable processing Flexicast QUIC on multiple threads is in the asynchronous module of the applications.
Do not forget to pull the submodules (boringssl in particular), otherwise the code won't compile!
git submodule update --init
The source historically relies on network coding, even if this is not supported anymore in the implementation. Normally, you should have access thanks to the access token. Do not hesitate to send us an email if this is not the case. At the time of writing, the token is still valid and works perfectly.
To compile the code, simply run
cargo build [--release] [--bin fc-server-rtp-tokio/fc-client-rtp]
We provide as much information as possible to reproduce the experiments from the paper. More information is available in the README from the experiments folder.
They rely on the Network Performance Framework (a Python module).
Please use the following bibtex input to cite this work:
@article{navarre2025taking,
title={Taking the Best of Multicast and Unicast with Flexicast QUIC},
author={Navarre, Louis and De Coninck, Quentin and Barbette, Tom and Bonaventure, Olivier},
journal={ACM SIGCOMM Computer Communication Review},
volume={55},
number={2},
pages={},
year={2025},
note={Postprint},
publisher={ACM New York, NY, USA}
}
quiche is an implementation of the QUIC transport protocol and HTTP/3 as specified by the IETF. It provides a low level API for processing QUIC packets and handling connection state. The application is responsible for providing I/O (e.g. sockets handling) as well as an event loop with support for timers.
For more information on how quiche came about and some insights into its design you can read a post on Cloudflare's blog that goes into some more detail.
quiche powers Cloudflare edge network's HTTP/3 support. The cloudflare-quic.com website can be used for testing and experimentation.
Android's DNS resolver uses quiche to implement DNS over HTTP/3.
quiche can be integrated into curl to provide support for HTTP/3.
quiche can be integrated into NGINX using an unofficial patch to provide support for HTTP/3.
Before diving into the quiche API, here are a few examples on how to use the quiche tools provided as part of the quiche-apps crate.
After cloning the project according to the command mentioned in the building section, the client can be run as follows:
$ cargo run --bin quiche-client -- https://cloudflare-quic.com/
while the server can be run as follows:
$ cargo run --bin quiche-server -- --cert apps/src/bin/cert.crt --key apps/src/bin/cert.key
(note that the certificate provided is self-signed and should not be used in production)
Use the --help
command-line flag to get a more detailed description of each
tool's options.
The first step in establishing a QUIC connection using quiche is creating a
Config
object:
let mut config = quiche::Config::new(quiche::PROTOCOL_VERSION)?;
config.set_application_protos(&[b"example-proto"]);
// Additional configuration specific to application and use case...
The Config
object controls important aspects of the QUIC connection such
as QUIC version, ALPN IDs, flow control, congestion control, idle timeout
and other properties or features.
QUIC is a general-purpose transport protocol and there are several configuration properties where there is no reasonable default value. For example, the permitted number of concurrent streams of any particular type is dependent on the application running over QUIC, and other use-case specific concerns.
quiche defaults several properties to zero, applications most likely need to set these to something else to satisfy their needs using the following:
set_initial_max_streams_bidi()
set_initial_max_streams_uni()
set_initial_max_data()
set_initial_max_stream_data_bidi_local()
set_initial_max_stream_data_bidi_remote()
set_initial_max_stream_data_uni()
Config
also holds TLS configuration. This can be changed by mutators on
the an existing object, or by constructing a TLS context manually and
<<<<<<< HEAD
creating a configuration using [with_boring_ssl_ctx()
].
creating a configuration using with_boring_ssl_ctx_builder()
.
multipath
A configuration object can be shared among multiple connections.
On the client-side the [connect()
] utility function can be used to create
a new connection, while [accept()
] is for servers:
// Client connection.
let conn = quiche::connect(Some(&server_name), &scid, local, peer, &mut config)?;
// Server connection.
let conn = quiche::accept(&scid, None, local, peer, &mut config)?;
Using the connection's [recv()
] method the application can process
incoming packets that belong to that connection from the network:
let to = socket.local_addr().unwrap();
loop {
let (read, from) = socket.recv_from(&mut buf).unwrap();
let recv_info = quiche::RecvInfo { from, to };
let read = match conn.recv(&mut buf[..read], recv_info) {
Ok(v) => v,
Err(e) => {
// An error occurred, handle it.
break;
},
};
}
Outgoing packet are generated using the connection's [send()
] method
instead:
loop {
let (write, send_info) = match conn.send(&mut out) {
Ok(v) => v,
Err(quiche::Error::Done) => {
// Done writing.
break;
},
Err(e) => {
// An error occurred, handle it.
break;
},
};
socket.send_to(&out[..write], &send_info.to).unwrap();
}
When packets are sent, the application is responsible for maintaining a
timer to react to time-based connection events. The timer expiration can be
obtained using the connection's [timeout()
] method.
let timeout = conn.timeout();
The application is responsible for providing a timer implementation, which
can be specific to the operating system or networking framework used. When
a timer expires, the connection's [on_timeout()
] method should be called,
after which additional packets might need to be sent on the network:
// Timeout expired, handle it.
conn.on_timeout();
// Send more packets as needed after timeout.
loop {
let (write, send_info) = match conn.send(&mut out) {
Ok(v) => v,
Err(quiche::Error::Done) => {
// Done writing.
break;
},
Err(e) => {
// An error occurred, handle it.
break;
},
};
socket.send_to(&out[..write], &send_info.to).unwrap();
}
It is recommended that applications pace sending of outgoing packets to avoid creating packet bursts that could cause short-term congestion and losses in the network.
quiche exposes pacing hints for outgoing packets through the [at
] field
of the [SendInfo
] structure that is returned by the [send()
] method.
This field represents the time when a specific packet should be sent into
the network.
Applications can use these hints by artificially delaying the sending of
packets through platform-specific mechanisms (such as the SO_TXTIME
socket option on Linux), or custom methods (for example by using user-space
timers).
After some back and forth, the connection will complete its handshake and will be ready for sending or receiving application data.
Data can be sent on a stream by using the [stream_send()
] method:
if conn.is_established() {
// Handshake completed, send some data on stream 0.
conn.stream_send(0, b"hello", true)?;
}
The application can check whether there are any readable streams by using
the connection's [readable()
] method, which returns an iterator over all
the streams that have outstanding data to read.
The [stream_recv()
] method can then be used to retrieve the application
data from the readable stream:
if conn.is_established() {
// Iterate over readable streams.
for stream_id in conn.readable() {
// Stream is readable, read until there's no more data.
while let Ok((read, fin)) = conn.stream_recv(stream_id, &mut buf) {
println!("Got {} bytes on stream {}", read, stream_id);
}
}
}
The quiche [HTTP/3 module] provides a high level API for sending and receiving HTTP requests and responses on top of the QUIC transport protocol.
<<<<<<< HEAD
[with_boring_ssl_ctx()
]: https://docs.quic.tech/quiche/struct.Config.html#method.with_boring_ssl_ctx
multipath [
connect()
]: https://docs.quic.tech/quiche/fn.connect.html [accept()
]: https://docs.quic.tech/quiche/fn.accept.html [recv()
]: https://docs.quic.tech/quiche/struct.Connection.html#method.recv [send()
]: https://docs.quic.tech/quiche/struct.Connection.html#method.send [timeout()
]: https://docs.quic.tech/quiche/struct.Connection.html#method.timeout [on_timeout()
]: https://docs.quic.tech/quiche/struct.Connection.html#method.on_timeout [stream_send()
]: https://docs.quic.tech/quiche/struct.Connection.html#method.stream_send [readable()
]: https://docs.quic.tech/quiche/struct.Connection.html#method.readable [stream_recv()
]: https://docs.quic.tech/quiche/struct.Connection.html#method.stream_recv [HTTP/3 module]: https://docs.quic.tech/quiche/h3/index.html
Have a look at the [quiche/examples/] directory for more complete examples on how to use the quiche API, including examples on how to use quiche in C/C++ applications (see below for more information).
quiche exposes a thin C API on top of the Rust API that can be used to more easily integrate quiche into C/C++ applications (as well as in other languages that allow calling C APIs via some form of FFI). The C API follows the same design of the Rust one, modulo the constraints imposed by the C language itself.
When running cargo build
, a static library called libquiche.a
will be
built automatically alongside the Rust one. This is fully stand-alone and can
be linked directly into C/C++ applications.
Note that in order to enable the FFI API, the ffi
feature must be enabled (it
is disabled by default), by passing --features ffi
to cargo
.
quiche requires Rust 1.66 or later to build. The latest stable Rust release can be installed using rustup.
Once the Rust build environment is setup, the quiche source code can be fetched using git:
$ git clone --recursive https://github.com/cloudflare/quiche
and then built using cargo:
$ cargo build --examples
cargo can also be used to run the testsuite:
$ cargo test
Note that BoringSSL, which is used to implement QUIC's cryptographic handshake
based on TLS, needs to be built and linked to quiche. This is done automatically
when building quiche using cargo, but requires the cmake
command to be
available during the build process. On Windows you also need
NASM. The official BoringSSL
documentation has
more details.
In alternative you can use your own custom build of BoringSSL by configuring
the BoringSSL directory with the QUICHE_BSSL_PATH
environment variable:
$ QUICHE_BSSL_PATH="/path/to/boringssl" cargo build --examples
Building quiche for Android (NDK version 19 or higher, 21 recommended), can be done using cargo-ndk (v2.0 or later).
First the Android NDK needs to be installed, either using Android Studio or
directly, and the ANDROID_NDK_HOME
environment variable needs to be set to the
NDK installation path, e.g.:
$ export ANDROID_NDK_HOME=/usr/local/share/android-ndk
Then the Rust toolchain for the Android architectures needed can be installed as follows:
$ rustup target add aarch64-linux-android armv7-linux-androideabi i686-linux-android x86_64-linux-android
Note that the minimum API level is 21 for all target architectures.
cargo-ndk (v2.0 or later) also needs to be installed:
$ cargo install cargo-ndk
Finally the quiche library can be built using the following procedure. Note that
the -t <architecture>
and -p <NDK version>
options are mandatory.
$ cargo ndk -t arm64-v8a -p 21 -- build --features ffi
See build_android_ndk19.sh for more information.
To build quiche for iOS, you need the following:
- Install Xcode command-line tools. You can install them with Xcode or with the following command:
$ xcode-select --install
- Install the Rust toolchain for iOS architectures:
$ rustup target add aarch64-apple-ios x86_64-apple-ios
- Install
cargo-lipo
:
$ cargo install cargo-lipo
To build libquiche, run the following command:
$ cargo lipo --features ffi
or
$ cargo lipo --features ffi --release
iOS build is tested in Xcode 10.1 and Xcode 11.2.
In order to build the Docker images, simply run the following command:
$ make docker-build
You can find the quiche Docker images on the following Docker Hub repositories:
The latest
tag will be updated whenever quiche master branch updates.
cloudflare/quiche
Provides a server and client installed in /usr/local/bin.
cloudflare/quiche-qns
Provides the script to test quiche within the quic-interop-runner.
Copyright (C) 2018-2019, Cloudflare, Inc.
See COPYING for the license.