MLS is well suited for large groups. Marmot uses MLS, but that doesn’t mean it scales in the same way, since some of the protocol’s conventions are shifted. Specifically, MLS requires strong coordination among all group members. This means that things like out-of-order messages, degraded network states, and similar issues break MLS. That’s why MLS deployments typically require a coordinator or delivery service to ensure group messages follow the strict ordering MLS demands.
Marmot, however, uses Nostr relays as both a dissemination layer and storage. Since there’s no delivery service or nothing ensuring that all members receive messages in the same order, all the coordination MLS requires must be handled client-side by each participant. Each participant must reach the same state to maintain group integrity. This means the complexity of keeping everyone on the same page scales with the group size.
For example, a group of two usually works without issues. A group of three requires that each member consistently lands in the same state, which is manageable under good hygiene and network conditions, meaning all relays in a set are online at all times, with no race conditions (e.g., one relay going offline while others don’t, you know classic nostr race conditions). But still three members can work under good conditions. But as you add more people, each new member increases the complexity of keeping everyone synchronized.
Anyway, sorry for the length 😅 I tried to explain myself as clearly as possible.
But in summary, MLS scales well for large groups but requires coordination. Without coordination, these scalability assumptions break down. I reached these conclusions after reading
@Gzuuus ’s article on private communications over public infrastructure, as well as discussions with him. It was very helpful indeed
I hope this clarifies things! Please don’t hesitate to ask if you have any questions, these are complex technical details :)
