access workers, routers, transports objects from one node process to another

Interesting, that picture cleared everything. Nice discussion thanks for your time brother, beautifully explained, it will help many.

No problem

You can also make use of my project Mafalda SFU, that’s built on top of Mediasoup just for the purpose of provide vertical and horizontal scalability without needing to change application lógic.

Nice work, will check it out.

1 Like

@piranna You’ve done what exactly with the project in the last year+? See no scaling or anything that insists the demo be a better trial.

I have not been publicly working on It during the last months due to an agreement with my employer https://dyte.io, since we have pivoted our business model and now Mafalda SFU is somewhat competence, but in personal tests, I’ve been able to make EduMeet to use a remote instance of Mediasoup with just 20 lines of code, that’s a keystone to have transparent horizontal scalability.

Main issue is about orchestrating, sharing and updating all the Mediasoup state over the network and among mĂșltiple instances, but more specially, appData objectos has been a real PITA.

Does dyte.io uses mediasoup internally or it has its own custom sfu implementation?

We were using Mediasoup, but since begin of year, we have been moving to our own in-house SFU.

Great? Mediasoup wasn’t enough for your usecases? And your custom sfu is written in c++?

Mediasoup is really good, but CTO decided to move to custom in-house SFU due to both being Mediasoup development, maintenance and roadmap too much centralized in a reduced development team, to optimize the SFU to our own scalability needs, and because most of our backend developers know Go and doesn’t know so much C++. Definitely, I would have not picked that decission.

Sounds like a waste of time, no performance gain and an obvious sign of coders lacking experience. I would hope with time-lost it is more efficient, otherwise what the heck?