Matrix chat rooms don't scale well

[pcppcp]

Result of the stress testing of Matrix server revealed that with increasing number of participants in the channel message delivery time increases exponentially. That's a problem for our proposed broadcast channel where Balance Proofs are broadcasted and Monitoring servers actively listen for messages sent to the channel.

Ideas:

• submit BP via direct message to MS. This would require some sort of MS discovery (can be done in a channel)
• split broadcast channels and use naming schema i.e. <chain_id>_<token_network_address>

TODOS:

• [ ] decide on how to avoid using broadcast rooms (use service discovery?)
• [ ] create metrics for proposed protocol
• [ ] decide which naming schema for the broadcast channels will be used

[heikoheiko]

Other solutions (with limited client implementation complexity):

• Clients not polling (calling GET) on the broadcast channel
• Each client sending to a dedicated room where only a homserver agent is listening who'd copy the message to the broadcast recipient channel.

I'd defer solving this issue as long as we are confident, that it can be solved at all - and go with a single broadcast channel for now.