The PSYC architecture is made of three parts:
MMP, the Conferencing Module and the technical PSYC itself.
MMP is the Modular Message Protocol,
it's a simple carrier protocol for the PSYC packets which therefore do
not need to get parsed and understood by simple MMP routers and proxies.
The Conferencing Module is a particularely elaborate approach in steering
chatrooms and online conferences in general - it is an extension to MMP.
PSYC itself operates end-to-end, leaving routing and multicast issues
to the lower layers.
PSYC and MMP share a common syntax, which looks unusual
compared to traditional text-based protocols, but has advantages in its
abilities to get compressed and to be semantically extended.
It is important to understand, that the architecture is totally different.
You could at best see it as something in-between IRC and XMPP, but that's
still quite a simplification.
servers dictate how users and groups want to be treated, from then on
clients communicate with each other according to the rules set up by
the servers, optionally using routers, proxies or complex multicast
infrastructures as transport media.
This is why PSYC conference control can be very well applied to
audio/video conferences, too. The actual data is still spread using
the appropriate realtime protocols.
The current real situation has servers relaying for their users a lot,
however. But that doesn't mean we won't become more peer-to-peer-capable
later on. The protocol is prepared for that step.
There is no central database whatsoever, the system works in a
worldwide distributed fashion like the web.