When a MH moves from one network to another, there may be an interval during which it has poor connectivity (either lost packets or low bandwidth) in the new network, but good connectivity in the old network. If the MH performs handoff too early, then its performance can suffer from poor connectivity in the new network. On the other hand, if the MH performs handoff too late, then it may lose packets as the connectivity in the old network degrades.
The solution in ROAM is to use the generalized level of indirection
provided by 
to do multicast-based soft handoff. In the situation
described above, when the MH can obtain an address in the new network,
the MH's proxy inserts a trigger with the same identifier as its
existing trigger, but associated with the new address. This causes the
same packets to be delivered to both the old and new addresses. This
allows the MH to take advantage of the best available connectivity.
There are two things worth noting. First, the use of multicast is
completely transparent to the sender. Second, fast handoff is still
necessary for cases when the MH cannot listen simultaneously at both
addresses. For example, an 802.11b client cannot be simultaneously
connected to two base stations on different channels
[27]. We address the problems of determining when to
stop using multicast and how to suppress duplicate packets in
Section 6.1. We discuss the implication of
multicast on communication privacy in Section 8.