* Both MOSPF and DVMRP do "pruning," which means to restrict the extent of multicast datagrams to those parts of an internetwork that actually have active group members.
* Since a DVMRP router does not know exactly where all the group members are, unlike MOSPF, DVMRP cannot optimize IP multicast's "expanding ring search" behavior (see earlier discussion on path characteristics).
These brouters would be acting as IBM source routing bridges on their local token ring segments, and as MOSPF routers over their WAN connections (e.g., over collections of synchronous serial lines, an X.25 PDN or a Frame relay network).
Using MOSPF, instead of a separate packet being sent to each brouter, a single packet can be sent to a well-known multicast address.
Such a system has been deployed using the Proteon MOSPF implementation.
MOSPF uses the Dijkstra algorithm to calculate the path of a multicast datagram through any given OSPF area.
MOSPF deals with this calculation burden by calculating datagram paths in an "on demand" fashion.
Other considerations when evaluating the cost of MOSPF's routing calculation are:
* A datagram whose destination has no group members in the Autonomous System can still be forwarded through the MOSPF system.
One other factor to be considered in MOSPF scaling is how often cache entries need to be recalculated, as a result of a network topology change.
The MOSPF protocol was designed in the MOSPF Working Group of the Internet Engineering Task Force.
1 The cases where the paths are not the same are when MOSPF routers are mixed with non-MOSPF routers, or when datagrams cross OSPF areas and/or Autonomous Systems with asymmetric link costs.