A broadcast is a special type of multicast; the addressing is such (conventionally all 1's) that every station that "hears" it will accept it as traffic meant for that station.
Once the frame as been accepted by that station, it goes up the stack (as in OSI model-type stack) and each level determines if the traffic really is for that station or not ... if it is (or can't yet be determined) then it goes up to the next level ... if it's not, the information is discarded.
Multicasts are similar, in that it is a one-to-many type of transmission. The difference is that multicasts have addressing such that they can be directed towards a group ... stations that are listening for traffic with that destination address (L3/IP addresses in the range of 224.0.0.0 to 239.255.255.255).
Where a broadcast must be accepted and acted on (to some degree) by the receiving station, a multicast is only accepted by stations that are listening for the specific group address, so it wastes less processing time on stations that don't care about the traffic.
A unicast is a one-to-one transmission. A specific sending station is sending traffic to a single specific destination station.
Multicasts are a bandwidth-saving mechanism (used properly). If you had 100 stations that wanted to view a 2 megabit mpeg2 video (for example), and you tried to unicast it, there would be several problems:
1) the sending station would have to form and send the same chunk of information 100 times (100 unicasts with specific source and destination addresses)
2) you'd need ~200+ megabits of bandwidth (100 stations * 2 meg stream) for a real-time presentation.
With a multicast, the sender addresses the traffic with a specific group address (like 239.1.1.10) and the 100 receiving stations would be listening for traffic on that address.
With multicast, the sender only has to form a single output stream (239.1.1.10), it it will only use the 2 megabits of bandwidth (single stream * 2 megabits).
The layer three address (239.1.1.10) is translated to a specific group of layer 2 (MAC)addresses (also designated as multicast addresses) and applied to the frame as the destination address. Stations that "aren't interested" in "hearing" that group of addresses will ignore that traffic, saving some stack and processor time for those stations (compared to a broadcast, where every station must accept the frame and evaluate the contents).
The last type, that you didn't ask about, but is frequently misunderstood, is a "flood."
When a bridge or a switch (really the same thing) hears a destination address that it hasn't heard before, it "floods" the frame out all ports (except the one it was heard on) to see it any of the stations connected to it are the recipient.
Some people describe it as "the switch BROADCASTS the frame out all ports " .... which is wrong. A broadcast (at layer two), by definition, has all ones in it's destination address .... with a flood, the source and destination addresses are intact (the same as what entered the switch/bridge), but the switch "floods" the frame out every port (except the one it was received on).
So, to summerize:
Unicast - single source sending to a single destination
Broadcast - single source sending to EVERY station in the broadcast domain
Multicast - single source sending to a specific group of stations (only in that broadcast domain, unless the routers are configured to propagate the multicast to other branches/networks/subnetworks)
Flood - Single station sending to a single destination, the receiving switch/bridge doesn't recognize the destination, so it sends the frame out all ports except the one from which the frame was received.
Hope this helps
Scott
