Mafalda SFU
Mafalda SFU is a suite of Node.js packages that provides both vertical and horizontal scaling on top of Mediasoup, at the same time they provide a familiar and easy to use API.
Features
- Automatically transparent abstracts and manage Mediasoup Workers and Routers, also located on different servers
- Designed for optimization of resources usage
- 100% tests code coverage of all projects for both lines, branches and statements
- Same orthogonal API for all modules inspired in Mediasoup design, easy to upgrade and combine between them
Packages (Mafalda sub-projects)
- Mafalda: Mediasoup vertical scalability, allow to have Router objects that by-pass Mediasoup per-Worker limits.
- Remote Mediasoup server & Remote Mediasoup client: remote control of Mediasoup instances.
- Remote Mafalda server & Remote Mafalda client: remote control of Mafalda instances.
- Mediasoup-horizontal: Mediasoup horizontal scalability, allow to use Mediasoup Worker instances located on multiple servers from a single place as if they were local to it.
- Mediasoup-cluster: standalone server using Mediasoup-horizontal to control multiple Remote Mediasoup servers from a single endpoint.
- Mafalda-horizontal: Mafalda horizontal scalability, allow to have Router objects that transparently span over multiple servers and CPUs. Currently in development
- Mafalda-swarm: Decentralized horizontal scaling for Mafalda, allow to access streams from multiple entry points without a central management, based on P2P technologies and architecture. Estimated September 2022
- Mafalda-monitor: Mafalda network monitoring tool, allow to control stats and loads of the servers in the network and their connected clients. Estimated December 2022
What package should I use?
Mafalda modular design allows you to use the packages that better fit for your use case, and easily upgrade to other ones when you need it. To help you to decide were to start, we’ve defined here some use cases:
I’m writting a new server
If you are writtin a new server from scratch, the best option is to start with Mafalda package. This one will allow you not only to abstract from creating and managing yourself the usage of multiple Mediasoup Workers, but also to load balance the creation Mediasoup Routers on them, or pipe and join the Router instance between them to by-pass Mediasoup per-Worker limits.
I’m using only non-interconnected Router instances that I need them to scale
If you already have an app server using Mediasoup Routers and until now you didn’t need them to interconnect them to by-pass Mediasoup per-Worker limits, it would pay-off the efforts of migrate your Mediasoup code to use Mafalda instead. Mafalda API and design is based on Mediasoup one, so migration should be easy to do, and would enjoy all the advantages of scalability and abstracted management that Mafalda provides, without almost needing to change your application logic.
In addition to that, by using Mafalda will prepare your code to later upgrade to use Mafalda-horizontal and scale your app to multiple servers.
I have mixed app logic and media, and want to split to improve architecture
If your app server code directly access Mediasoup API, and need to move it out to another machine to improve your systems architecture or use a server with more resources to improve media performance, the easiest solution is to use Remote Mediasoup package. After creating and configuring a Remote Mediasoup client, it will expose an object with the same API that Mediasoup provides, so your you’ll be able to control a Mediasoup instance running in another server without needing to modify the logic of your app code at all.
Alternatively, you can also use Mediasoup-horizontal, that will already prepare your application to scale to multiple servers without changes, or use a Remote Mediasoup client instance to connect to a Mediasoup-cluster server.
I’m using Mafalda, and want to move out media server to improve performance
If you are already using Mafalda in a monolithic server, you can split your code and use a remote media server by using Remote Mafalda package and improve your performance by having a dedicated media server. Similar to Remote Mediasoup, you’ll be able to control a Mafalda instance running in another server, and prepare your code to upgrade later to use Mafalda-horizontal and allow each one of your sessions to grow and span over multiple CPUs and servers.
My app uses Mediasoup, and I need to scale it to a large number of servers
If you need your Mediasoup-based app to being able to manage multiple media servers, your solutions are to use Mediasoup-horizontal if you want your app to control them directly, or use Mediasoup-cluster if you want to split the management of the multiple media servers from your actual app logic. Both of them will allows you to control multiple Mediasoup instances as a single one, and to have remote Workers and Routers all over them in a transparent way, as if running a single machine with a lot of CPU cores.
I need to host huge sessions that will span several CPUs in multiple servers
If your sessions can be potentially big and exceed the capability that can offer a single server, Mafalda-horizontal will help here. It will automatically create, manage and interconnect the Router instances on the different servers while providing the same API of Mafalda, so you’ll only need to worry about using it in your app code as if you only would need a single Router local instance.
Currently in development
I host humongous big sessions, and need to access them from multiple places
If your sessions can be really big or can’t be predicted when that usage spikes could happen or can be a lot of difference in sessions load from ones to other, if the number of servers you need to manage is huge or you can’t easily control when or how to add them, if you need that stream distribution auto regulates itself finding the shortest path, or if you need to access the servers network from multiple entry points (for example, from multiple regions), then now we’re talking.
Mafalda-swarm will provide a federated and decentralized P2P architecture build on top of the Mafalda network, self-managing its resources and allowing access to it from any of their nodes by providing a search mechanism of the streams.
Estimated September 2022
I need to monitor the stats, loads and connections of all my Mafalda servers
Not directly related to solve scalability, but scalability-derived ones, Mafalda-monitor allows to control and monitor from a single place the activity of all your Remote Mafalda servers and their network clients. With it, you can have a real-time graphical view of the number of their Workers and Routers, their CPU loads, their Transports, how are they inter-connected…
Estimated December 2022