From the backend of it (I was on the team for it) Even with our "light" load, Matrix/Riot is a overbloated product overall. To handle the scale of Defcon would require hardware I don't think exists

I ran the Matrix server for HOPE and have fairly extensive experience administrating Matrix. For HOPE, we didn't even begin to approach anything that looked like capacity; any resources that were maxing out were either due to the DoS attack we were enduring (and made numerous changes to mitigate) during the conf, or applications in different threads that matrix metrics were picking up on (bug filed, will fixed in new versions).

Even so, we could easily - without a sweat - get 2000 people on a server. I wasn't even using more workers than we had cores. Make that a bigger box and you get a few thousand more. Then use a few different servers that are federated with each other for DEFCON and you have instant scale.

The matrix.org people run the matrix.org homeserver for over ~200k active users. You can see an outdated version of the stats here: https://matrix.org/faq/#what-is-the-current-project-status and they've said their active userbase has grown 300% in the past year.

The original plan for HOPE was to run multiple homeservers, but we didn't even begin to need to.

Also soonish, we'll have dendrite (go) and conduit (rust).

EDIT: And eventually p2p which might change everything.

The matrix.org server deployment handles around 100K concurrent users these days across ~4M rooms, with ~30Hz of messages in and ~3000Hz of messages out - and these days has average traffic send time under 100ms.

DEFCON should be trivial by comparison, and as others have said, the HOPE server (2500 users or so) was a resounding success - so much so that it’s been set up as permenant fixture now post conference.

Now, it’s true that the Synapse server implementation has to be tuned - for large userbases you need worker processors and to dial up the caches a bit. And needless to say you need postgres rather than sqlite.

Kinda depressing that nobody from DEFCON synced with the Matrix team (to my knowledge) on how best to do this, rather than just writing it off as slow and bloated.