Many Listeners

IP networks have supported multicast transmission for nearly 20 years, but the technology has only recently entered the realm of widespread use. As the name implies, multicasting is a technique for transmitting data from a single source to a predefined collection of recipients. This concept poses some special challenges that aren't seen in more conventional transmission techniques, such as broadcasting, in which the message is sent to all computers on a network segment, and unicasting, in which a message passes from a single source to a single recipient.

Efficient use of multicasting can significantly reduce traffic load, especially on networks that support streaming-style multimedia transmissions. Multicasting applications and technologies have received increased attention with the rise of audiovisual technologies; however, multicasting remains a mystery to many software developers, system admins, and end users who might benefit from more extensive use of this promising technique. In this article, we offer a glimpse at the practical side of multicasting, including a sample configuration that uses the free XORP routing protocol suite.

What Is IP Multicasting?

Figure 1 shows the idea behind multicast transmission. Source A generates a data stream with throughput of 1Mbps, and the data stream is received by three recipients. Figure 1a shows a unicast transmission between the source and the receivers. The transmission results in three independent, but identical, data streams, which means that a bandwidth of 3Mbps is consumed on the link between the source and distribution networks. In contrast, a multicast scenario (see Figure 1b) requires only one data stream from the source, so the load on the link is constant and independent of the number of receivers.

Figure 1a: Unicast transmission.

Figure 1b: Multicast transmission.

Data Link Layer

Transmission at the Data Link Layer is performed through the use of the MAC address, which identifies a network interface on the link. MAC addresses are mapped to and from the corresponding IP addresses with the help of the address resolution protocol (ARP) and reverse ARP (RARP). For example, if router B wants to send data to router D, it first sends an ARP request to D's IP address. In response, D sends an ARP reply that contains its MAC address. Once this process is finished, B and D can communicate over the Ethernet transmission medium. In the case of multicasting, the question is how to achieve the effect of Data Link addressing without incurring the overhead of having to resolve an IP address to an arbitrarily complex tangle of recipient MAC addresses.

This particular problem is solved by mapping an IP multicast address to a single MAC address that is then used by all recipients. An ethernet MAC address consists of 48 bits. Ethernet addresses that start with the bit sequence 01.00.5E are assigned to IANA (Internet Assigned Number Authority, the organization responsible for managing IP address ranges). IANA has decided to allocate half of the assigned Ethernet address range to the purpose of multicast transmission. As a result, 23 bits of the MAC address are available for the purpose of group communications.

An IP address, however, has 32 bits, which means 32 bits of the IP address must map to 23 bits of the MAC address. All IP addresses from Class D are reserved for multicast transmission purposes. IP addresses belonging to Class D start with the bit pattern 1110. Because this pattern is constant for all multicast addresses, it doesn't have to be part of the mapping. Consequently, only 28 bits of the IP address are mapped to 23 bits of the MAC address representation. Figure 2 shows how the mapping procedure works.

Figure 2: Mapping an IP multicast address to a MAC address.

As you can see in Figure 2, after discarding the initial 4 bits, the next 5 most-significant bits are also ignored. The remaining 23 bits are then directly mapped to the MAC address. Note that this mapping of IP to MAC multicast addresses is not unambiguous; the relation between IP and MAC addresses is not one-to-one. Consequently, 25 IP group addresses have the same MAC addresses.

A simple example will illustrate this process. Consider the multicast address 239.16.16.46, which has the following binary representation: 11101111.00010000.00010000.00101110.

Discard the 4 most-significant bits (the Class D bit pattern) and you have the following sequence: 1111.00010000.00010000.00101110.

If you omit the next 5 bits, for 0010000.00010000.00101110, and combine this bit pattern with the sequence assigned by IANA for the purpose of multicast transmissions, you can compute the MAC address 01.00.5E.10.10.2E corresponding to the IP address 239.16.16.46.

How It Works

When an application requests reception of multicast transmission, the kernel network subsystem computes the corresponding MAC address. This address is then added to the list of received multicast addresses; finally, the kernel calls set_multicast_list from the net_device structure. The set_multicast_list function performs some hardware-specific actions at the driver level so the network card will be able to receive packages sent to this specific MAC address.

Figure 3 shows an example of a multicast transmission taking place between two systems connected to the same Ethernet network segment. The application on computer A transmits data to multicast address 239.16.16.46 (step 1a in Figure 3). The application passes the data, together with the destination address, to the kernel. The kernel then computes a MAC address corresponding to this IP address.

Figure 3: Multicast transmission scheme.

On the other end, the application on computer B informs the Linux kernel that it is interested in receiving a multicast transmission sent to IP address 239.16.16.46 (step 1b in Figure 3). The kernel computes the MAC address corresponding to this IP address (2b) and informs the network card that it should receive packages sent to this MAC address (3b). Once the data package is received (4b), the network card installed in computer B generates an interrupt, and the method responsible for interrupt handling is called (5b). This method hands over the received data to the kernel (6b), which will ultimately relay that data to the application (7b).