One trivial algorithm to make worldwide multicast traffic available everywhere could be to send it... everywhere, despite someone wants it or not. As this does not seem quite optimized, several routing algorithms and forwarding techniques have been implemented.
DVMRP (Distance Vector Multicast Routing Protocol) is, perhaps, the one most multicast routers use now. It is a dense mode routing protocol, that is, it performs well in environments with high bandwidth and densely distributed members. However, in sparse mode scenarios, it suffers from scalability problems.
Together with DVMRP we can find other dense mode routing protocols, such as MOSPF (Multicast Extensions to OSPF -Open Shortest Path First-) and PIM-DM (Protocol-Independent Multicast Dense Mode).
To perform routing in sparse mode environments, we have PIM-SM (Protocol Independent Multicast Sparse Mode) and CBT (Core Based Trees).
OSPF version 2 is explained in RFC 1583, and MOSPF in RFC 1584. PIM-SM and CBT specifications can be found in RFC 2117 and 2201, respectively.
All this routing protocols use some type of multicast forwarding, such as flooding, Reverse Path Broadcasting (RPB), Truncated Reverse Path Broadcasting (TRPB), Reverse Path Multicasting (RPM) or Shared Trees.
It would be too long to explain them here and, as short descriptions for them are publicly available, I'll just recommend reading the draft-ietf-mboned-in.txt text. You can find it in the same places RFCs are available, and it explains in some detail all the above techniques and policies.