IP Multicasting Issues

By Chris Koeritz

This document describes IP multicast and compares it with directed IP broadcast. It is intended as a brief introduction rather than a technical explanation. For more information than is available here, you might want to check out this link: http://www.multicasttech.com/faq/.

Directed IP Broadcast

The diagram below shows the essential points of a directed broadcast. The sender transmits some data (denoted by the little envelopes) out to the three subnets depicted. The data must actually be sent three times, one for each subnet. On the receiving subnets, every machine on the subnet will be sent the data, whether it wants it or not.

IP Multicast

The diagram below shows the essential points of a multicast. The sender transmits some data out to a multicast address. The data only needs to be sent once, and the routers will take care of the rest. The middle subnet has no hosts registered for this data and so will not see the data at all. On the other two subnets, only the hosts that have specifically registered for the data will see it; the other machines on the subnet will not. Note that the routers manage the delivery list; the sender need not know who the recipients are (and it will not know the recipients unless a higher level protocol is providing the information).

An Assortment of Multicast Issues

Multicast Group Addresses

IP multicast groups have a group address.
This is similar to an IP address but in a restricted range of 224.0.0.0 through 239.255.255.255.
These are called "Class D" addresses.
The range from 224.0.0.0 through 224.0.0.255 are reserved for multicast routing information.

Address Allocation

You can be officially assigned an IP multicast address by the MBONE (multicasting backbone) group. This indicates that you are the only authorized user of that particular address.
Multicast addresses can also be dynamically assigned. A registration process will provide a randomly chosen IP multicast group address. This choice of address must be communicated to all programs that require the broadcasts, potentially during a login process at a server.

Multicast Properties

IP multicast only supports sending packets of data, not arbitrary length messages.
Multicast is also intrinsically unreliable.
There are other higher level protocols that make multicast reliable, but none have been chosen as standards yet. (For example, PGM: Pragmatic General Multicast.)
IP multicast groups are called "open" because anyone can send datagrams to a group.

Multicast Scope

The scope of a multicast send is governed by the Time To Live (TTL) information in the data packet.
TTL is a number that dictates how many router hops the packet is allowed to make.
The classification scheme for TTLs is as follows:

0: restricted to same host; no other machines can see the data.
1: restricted to same subnet; just 'this' subnet will see the data.
32: restricted to same site; as dictated by network topology and firewalls.
64: restricted to same region; perhaps a country boundary.
128: restricted to same continent.

Comparison With Standard TCP/IP

Sending multicasts is easy. This is just a matter of sending to the IP multicast address, and specifying a port number on that address. This is essentially the same as sending a regular UDP (user datagram protocol) packet.
Receiving multicasts is not too hard either. One joins a multicast group by registering with the IP multicast address. The address doesn't correspond to a specific machine, but to a virtual group defined by the IP address and the port number.
Once one has joined a multicast group, a TCP/IP socket can be opened. Socket usage is the same as for any other kind of TCP/IP communication. Now we will receive any data that is sent to the multicast group at our port number.

Comparison With Directed IP Broadcast

What are the drawbacks of using multicast?

Multicast is a newer protocol and needs support from a multicast enabled router.

Multicast transmissions devolve to a directed broadcast when there are no multicast routers, so even when customers use a multicast enabled application, they might actually be getting broadcasts on their lan.
Customers with older routers can still be supported with our existing directed broadcast mechanism

Multicast is an unreliable protocol.

There are no guarantees for delivery of transmissions.
Reliable multicast exists in several higher level protocols, but none of these are standard yet. These are also not supported by any routers yet, so they are implemented in software rather than hardware.
However, directed broadcast is also unreliable. Software can provide a reliable broadcast by supporting a retransmission process.

Multicast is wide open.

With directed broadcasts, the network administrator can configure our system however desired and know that traffic is not going off of the LAN/WAN.
Multicast group membership is open to machines across the internet and anyone can listen to multicast traffic if they know the address.
To enforce multicast security / visibility, one must either control access at the software level (by restricting the "time to live" on multicast packets, which could still expose the traffic to unwanted listeners) or at the hardware level (by modifying the router's multicast properties).

Why is multicast a potentially superior choice for data distribution?

With directed broadcast, one must know the recipients for the data in terms of their subnets. In IP Multicast, one does not need to know the recipients ahead of time.
In a directed broadcast, every machine on an entire subnet sees the data. With multicast, only the machines that have registered will see the data. The machines that are not included in a group will not see any traffic, whereas in a directed broadcast, each machine must discard the packets it was not interested in; thus multicast is less of a drain on computing resources.