About
What is Mafalda SFU?
Mafalda SFU is a suite of software packages that allows to create and manage WebRTC audio and video conference rooms and media streaming sessions with thousands of participants over multiple servers, in a fully automated way.
Built on top of Mediasoup, one of the most performant Open Source SFUs, Mafalda SFU allows to:
- by-pass Mediasoup by-design single-threaded limitations and provides full multi-threaded and multi-server support
- provides vertical and horizontal scalability in a transparent way for both apps code, users and developers
- manages itself the allocation and balancing of resources in the most optimal way
- and all of it, with the only limit of the provided hardware resources
You only need to focus on your application logic, and Mafalda SFU will take care of management of the media scalability over all your CPUs and servers for you. Following development best practices, code is split in multiple modules with a common API, allowing to combine between them in the most optimal way for your specific use case.
Who is Mafalda SFU for?
Since idea conception in March 2021, started development as a proof of concept in April 2021, and first public comments about its development in May, June and July of 2021, Mafalda SFU has been receiving interest from companies on different sectors, like:
- Development of videoconferencing platforms and SDKs
- Videoconferencing and audio/video applications with high number of concurrent rooms or participants
- Low latency media streaming and distribution applications
- 2D/3D/VR metaverses & social platforms
- Multi-source 3D audio applications
- Online real-time education platforms
Mafalda SFU also allows to scale current Mediasoup based videoconference applications without needing to change their application logic. Just by replacing Mediasoup with Mafalda SFU, and configure and initialize the library, you can scale your application to thousands of participants in a matter of minutes.
What Mafalda SFU scalability solution better suits me?
Mafalda SFU modularity makes it ery flexible regarding use cases or architecture designs, but we have had a set of common use cases we have seen when working on other previous projects. That doesn’t means that Mafalda SFU would not be able to work in other scenarios, nor we could not be able to adapt it to your use case on demand, but theses ones show how much versatile Mafalda SFU can be.
Lots of small sized rooms (each one under a single CPU capacity limits)?
Horizontal scalability, also known as scale out: Mediasoup Cluster controlling several Remote Mediasoup server instances running on top of vanilla Mediasoup. Alternatively, Mediasoup-horizontal can be used standalone (running in local) to control the Remote Mediasoup server instances, but it’s not recommended for performance and stability reasons.
One medium sized room (under capacity limits of multi-CPUs single server)?
Vertical scalability, also known as scale up: Mediasoup Vertical running on top of vanilla Mediasoup, aggregating all the systems resources and offering them to the application as if running on a single, huge CPU.
Some medium sized rooms (under capacity limits of each single server)?
Cluster of Mediasoup Verticals (horizontal-over-vertical scalability): Mediasoup Cluster controlling several Remote Mediasoup server instances running on top of Mediasoup-Vertical. This is not provided as a standalone product, but instead it’s offered as an architecture configuration option.
One or a few of big sized rooms (expanding over multiple servers)?
Mediasoup vertical cluster (vertical-over-horizontal scalability, also known as aggregated scalability): Mediasoup-vertical running on top of Mediasoup-horizontal, controlling several Remote Mediasoup server instances running on top of vanilla Mediasoup. For performance reasons, it’s encouraged the Remote Mediasoup server instances make use of the Mediasoup Vertical module themselves (for example, being Remote Mediasoup vertical server, Remote vertical proxy, or Remote vertical cluster instances), to improve closer locality of related Mediasoup Workers in a room.
Mafalda SFU ❤️ Open Source
Mafalda SFU is build on top of multiple Open Source projects, being Mediasoup the most prominent one. In addition to that, during development of Mafalda SFU we have created and published multiple auxiliary Open Source projects and contributed to several others, between dependencies, tools, demos, reporters, forks… You can find a list of all our Open Source repositories and contribution forks at Mafalda SFU Github organization, at Mafalda SFU NPM organization, and a list of Mafalda SFU team created issues, PRs & Open Source contributions at Mafalda SFU Notion page.
A Jesús Leganés-Combarro project
With more than 20 years of experience in software development, and being one of Spain first WebRTC experts and pioneers since 2012 when he developed ShareIt!, the World’s first in-browser P2P file sharing application, and DataChannel-polyfill, the first working implementation of WebRTC DataChannel specification, Jesús Leganés-Combarro has been working on development of real-time and high performant applications for the last 16 years for companies like Kurento (now part of Twilio), Telefónica, Atos, IE University, Dyte, or TRC and Spain’s army. He has also worked hand-to-hand with other world-renowned WebRTC experts and pioneers like Iñaki Baz Castillo (creator of Mediasoup itself), Sergio Garcia Murillo, Professor Luis Lopez Fernandez, or Tsahi Levent-Levi, among others.
FAQ
Where does the name of Mafalda SFU come from?
Name of the project has been taken as a tribute to Mafalda, the character created by Joaquín Salvador Lavado Tejón ‘Quino’, that has a love-hate relationship with both mass media and soup 😉
What’s the Mafalda SFU projects code coverage?
Most of the main dependencies and auxiliary projects have more than 80% code coverage (in some cases that gets up to 100% code coverage for both statements, branches and functions). For all main and core Mafalda SFU projects, code coverage gets up to more than 90%.
Will Mafalda SFU allows me to display thousands of participants in a single room?
TL;DR: Not by itself, but we can help you to achieve it.
In a similar way to how Mediasoup does, Mafalda SFU works like a library that allows to create and manage WebRTC audio and video conference rooms and media streaming sessions with thousands of participants at the same time over multiple servers, in a fully automated way. The bottleneck in this case is the number of participants that can be played at the same time, that’s not a limitation of Mediasoup or Mafalda SFU, but of the browser or mobile app clients that are playing them, and the capacity of the computer or mobile phone they are running on top of. Several tests has shown that Chrome browser in a regular Intel i7 laptop can smoothly play up to 16 videos at the same time, starting to be unstable with 20-25 videos, and being practically useless with 30 or more videos, also with low resolution and framerate. On mobile phones and tablets devices, it’s not practical to show more than 6 videos a the same time, mostly due to screen space available.
In addition to that practical hardware or bandwidth limits from the clients side, Chrome itself imposed a hard limit of 200 tracks (audio or video) on desktop and 100 tracks on mobile due to these quality issues, although it was later uplift to 1000 tracks until a better solution was implemented since there was critics from developers of high demanding web applications, but as of January 2024 it’s still not implemented.
If your application needs to play more streams than the browser or mobile app can handle, you can use Mafalda SFU to manage the media streams and distribute them to the clients, but you’ll need to implement your own logic to display them in a way that is useful for your application. For example, in a video conference app you can have a layout where only the most recently active users are being shown, and push back others to a list, so you can mute or stop their video tracks and only play the audio of that less recently active participants, or in a metaverse you can play the audio of the closest avatars, saving both CPU processing and bandwidth at the same time.
Mafalda SFU team have been involved in the past in projects with similar requirements (that’s one of the reasons we started Mafalda SFU project on the first place), and we are capable of offer you consultancy services to help you define what’s the best architecture for your particular use case and needs, or also we can be able to implement it for you.