Border Gateway Protocol, Route Manipulation, and IP Multicast

IP Multicast Review

With multicast, packets are sent to a multicast group, which is identified with an IP multicast address. Multicast supports the transmission of IP packets from one source to multiple hosts. Packets with unicast addresses are sent to one device, and broadcast addresses are sent to all hosts; packets with multicast addresses are sent to a group of hosts.

Multicast Addresses

Multicast addressing uses Class D addresses from the IPv4 protocol. Class D addresses range from 224.0.0.0 to 239.255.255.255. IANA manages multicast addresses.

Routing protocols (RIPv2, EIGRP, and OSPF) use multicast addresses to speak to their neighbors. For example, OSPF routers use 224.0.0.6 to speak to the designated router (DR) in a multiaccess network. Class D multicast addresses range from 224.0.0.0 to 239.255.255.255. Multicast addresses in the range of 224.0.0.1 to 224.255.255.255 are reserved for special addresses or network protocol on a multiaccess link. RFC 2365 reserves multicast addresses in the range of 239.192.000.000 to 239.251.255.255 for organization-local scope. Similarly, 239.252.000.000 to 239.252.255.255, 239.254.000.000 to 239.254.255.255, and 239.255.000.000 to 239.255.255.255 are reserved for site-local scope.

Table 12-2 lists some well-known and multicast address blocks.

Table 12-2. Multicast Addresses

Layer 3 to Layer 2 Mapping

Multicast-aware Ethernet, Token Ring, and Fiber Distributed Data Interface (FDDI) network interface cards use the reserved IEEE 802 address 0100.5e00.0000 for multicast addresses at the MAC layer. This includes Fast Ethernet and Gigabit Ethernet. Notice that for the address, the high-order byte 0x01 has the low-order bit set to 1. This bit is the Individual/Group (I/G) bit. It signifies whether the address is an individual address (0) or a group address (1). Hence, for multicast addresses, this bit is set to 1.

Ethernet interfaces map the lower 23 bits of the IP multicast address to the lower 23 bits of the MAC address 0100.5e00.0000. As an example, the IP multicast address 224.0.0.2 is mapped to the MAC layer as 0100.5e00.0002. Figure 12-15 shows another example looking at the bits of multicast IP 239.192.44.56. The IP address in hexadecimal is EF:C0:2C:38. The lower 23 bits get mapped into the lower 23 bits of the base multicast MAC to produce the multicast MAC address 01:00:5E:40:2C:38.

Figure 12-15 Mapping of Multicast IP Addressing to MAC Addresses

IGMP

IGMP is the protocol used in multicast implementations between the end hosts and the local router. RFC 2236 describes IGMP Version 2 (IGMPv2). RFC 3376 describes IGMP Version 3 (IGMPv3). RFC 1112 describes the first version of IGMP.

IP hosts use IGMP to report their multicast group memberships to routers. IGMP messages use IP protocol number 2. IGMP messages are limited to the local interface and are not routed.

IGMPv1

The first RFC describing IGMP (RFC 1112), written in 1989, describes the host extensions for IP multicasting. IGMPv1 provides simple message types for communication between hosts and routers. These messages are

• Membership query—Sent by the router to check whether a host wants to join a multicast group
• Membership report—Sent by the host to join a multicast group in the segment

The problem with IGMPv1 is the latency involved for a host to leave a group. With IGMPv1, the router sends membership queries periodically; a host must wait for the membership-query message to leave a group. The query interval is 60 seconds, and it takes three query intervals (3 minutes) for a host to leave the group.

IGMPv2

IGMPv2 improves over IGMPv1 by allowing faster termination or leaving of multicast groups.

IGMPv2 has three message types, plus one for backward compatibility:

• Membership query—Sent by the router to check whether a host wants to join a group.
• Version 2 membership report—A message sent to the group address with the multicast group members (IP addresses). It is sent to by hosts to join and remain in multicast groups on the segment.
• Version 2 leave group—Sent by the hosts to indicate that a host will leave a group; it is sent to destination 224.0.0.2. After the host sends the leave group message, the router responds with a group-specific query.
• Version 1 membership report—For backward compatibility with IGMPv1 hosts.

You enable IGMP on an interface when you configure a multicast routing protocol, such as PIM. You can configure the interface for IGMPv1, IGMPv2 or IGMPv3.

IGMPv3

IGMPv3 provides the extensions required to support source-specific multicast (SSM). It is designed to be backward-compatible with both prior versions of IGMP.

IGMPv3 has two message types, plus three for backward compatibility:

• Membership query—Sent by the router to check that a host wants to join a group.
• Version 3 membership report—A message sent to the group address with the multicast group members (IP addresses). It is sent by hosts to request and remain in multicast groups on the segment.
• Version 2 membership report—A message sent to the group address with the multicast group members (IP addresses). It is sent by hosts to request and remain in multicast groups on the segment. This message is used for backward compatibility with IGMPv2 hosts.
• Version 2 leave group—Sent by the hosts to indicate that a host will leave a group, to destination 224.0.0.2. The message is sent without having to wait for the IGMPv2 membership report message. This message is used for backward compatibility with IGMPv2 hosts.
• Version 1 membership report—This message is used for backward compatibility with IGMPv1 hosts.

You enable IGMP on an interface when you enable a multicast routing protocol, such as PIM. You can configure the interface for IGMPv1, IGMPv2, or IGMPv3.

CGMP

CGMP is a Cisco-proprietary protocol implemented to control multicast traffic at Layer 2. Because a Layer 2 switch is unaware of Layer 3 IGMP messages, it cannot keep multicast packets from being sent to all ports.

As shown in Figure 12-16, with CGMP the LAN switch can speak with the IGMP router to find out the MAC addresses of the hosts that want to receive the multicast packets. With CGMP, switches distribute multicast sessions only to the switch ports that have group members.

Figure 12-16 CGMP

When a router receives an IGMP report, it processes the report and then sends a CGMP message to the switch. The switch can then forward the multicast messages to the port with the host receiving multicast traffic. CGMP fast-leave processing allows the switch to detect IGMP Version 2 leave messages sent by hosts on any of the switch ports. When a host sends the IGMPv2 leave message, the switch can then disable multicasting for the port.

IGMP Snooping

IGMP snooping is another way for switches to control multicast traffic at Layer 2. It listens to IGMP messages between the hosts and routers. If a host sends an IGMP query message to the router, the switch adds the host to the multicast group and permits that port to receive multicast traffic. The port is removed from multicast traffic if the host sends an IGMP leave message to the router. The disadvantage of IGMP snooping is that it has to process every IGMP control message, which can impact the CPU utilization of the switch.

Sparse Versus Dense Multicast Routing Protocols

IP multicast traffic for a particular (source, destination group) multicast pair is transmitted from the source to the receivers using a spanning tree from the source that connects all the hosts in the group. Each destination host registers itself as a member of interesting multicast groups through the use of IGMP. Routers keep track of these groups dynamically and build distribution trees that chart paths from each sender to all receivers. IP multicast routing protocols follow two approaches.

The first approach assumes that the multicast group members are densely distributed throughout the network (many of the subnets contain at least one group member) and that bandwidth is plentiful. The approach with dense multicast routing protocols is to flood the traffic throughout the network and then, at the request of receiving routers, stop the flow of traffic on branches of the network that have no members of the multicast group. Multicast routing protocols that follow this technique of flooding the network include DVMRP, Multicast Open Shortest Path First (MOSPF), and Protocol-Independent Multicast-Dense Mode (PIM-DM).

The second approach to multicast routing assumes that multicast group members are sparsely distributed throughout the network and that bandwidth is not necessarily widely available. Sparse mode does not imply that the group has few members, just that they are widely dispersed. The approach with sparse multicast routing protocols is to not send traffic until it is requested by the receiving routers or hosts. Multicast routing protocols of this type are Core-Based Trees (CBT) and Protocol-Independent Multicast-Sparse Mode (PIM-SM). CBT is not widely deployed and is not discussed in this book.

Multicast Source and Shared Trees

Multicast distribution trees control the path that multicast packets take to the destination hosts. The two types of distribution trees are source and shared. With source trees, the tree roots from the source of the multicast group and then expands throughout the network in spanning-tree fashion to the destination hosts. Source trees are also called shortest-path trees (SPT) because they create paths without having to go through a rendezvous point (RP). The drawback is that all routers through the path must use memory resources to maintain a list of all multicast groups. PIM-DM uses a source-based tree.

Shared trees create the distribution tree's root somewhere between the network's source and receivers. The root is called the RP. The tree is created from the RP in spanning-tree fashion with no loops. The advantage of shared trees is that they reduce the memory requirements of routers in the multicast network. The drawback is that initially the multicast packets might not take the best paths to the receivers because they need to pass through the RP. After the data stream begins to flow from sender to RP to receiver, the routers in the path optimize the path automatically to remove any unnecessary hops. The RP function consumes significant memory on the assigned router. PIM-SM uses an RP.

PIM

PIM comes in two flavors: sparse mode (PIM-SM) and dense mode (PIM-DM). The first uses shared trees and RPs to reach widely dispersed group members with reasonable protocol bandwidth efficiency. The second uses source trees and reverse path forwarding (RPF) to reach relatively close group members with reasonable processor and memory efficiency in the network devices of the distribution trees.

With RPF, received multicast packets are forwarded out all other interfaces, allowing the data stream to reach all segments. If no hosts are members of a multicast group on any of the router's attached or downstream subnets, the router sends a prune message up the distribution tree (the reverse path) to tell the upstream router not to send packets for the multicast group. So, the analogy for PIM-DM is the push method for sending junk mail, and the intermediate router must tell upstream devices to stop sending it.

PIM-SM

PIM-SM is defined in RFC 2362 (experimental). PIM-SM assumes that no hosts want to receive multicast traffic unless specifically requested. In PIM-SM, a router is selected as the RP. The RP gathers the information from senders and makes the information available to receivers. Routers with receivers have to register with the RP. The end-host systems request multicast group membership using IGMP with their local routers. The routers serving the end systems then register as traffic receivers with the RPs for the specified group in the multicast network.

Joining PIM-SM

With PIM-SM, DRs on end segments receive IGMP query messages from hosts wanting to join a multicast group. The router checks whether it is already receiving the group for another interface. If it is receiving the group, the router adds the new interface to the table and sends membership reports periodically on the new interface.

If the multicast group is not in the multicast table, the router adds the interface to the multicast table and sends a join message to the RP with multicast address 224.0.0.13 (all PIM routers) requesting the multicast group.

Pruning PIM-SM

When a PIM-SM does not have any more multicast receiving hosts or receiving routers out any of its interfaces, it sends a prune message to the RP. The prune message includes the group to be pruned or removed.

PIM DR

A designated router is selected in multiaccess segments running PIM. The PIM DR is responsible for sending join, prune, and register messages to the RP. The PIM router with the highest IP address is selected as the DR.

Auto-RP

Another way to configure the RP for the network is to have the RP announce its services to the PIM network. This process is called auto-RP. Candidate RPs send their announcements to RP mapping agents with multicast address 224.0.1.39 (cisco-rp-announce). RP mapping agents are also configured. In smaller networks, the RP can be the mapping agent. Configured RP mapping agents listen to the announcements. The RP mapping agent then selects the RP for a group based on the highest IP address of all the candidate RPs. The RP mapping agents then send RP-discovery messages to the rest of the PIM-SM routers in the internetwork with the selected RP-to-group mappings.

PIMv2 Bootstrap Router

Instead of using auto-RP, you can configure a PIMv2 bootstrap router (BSR) to automatically select an RP for the network. The RFC for PIM Version 2, RFC 2362, describes BSR. With BSR, you configure BSR candidates (C-BSR) with priorities from 0 to 255 and a BSR address. C-BSRs exchange bootstrap messages. Bootstrap messages are sent to multicast IP 224.0.0.13 (all PIM routers). If a C-BSR receives a bootstrap message, it compares it with its own. The largest priority C-BSR is selected as the BSR.

After the BSR is selected for the network, it collects a list of candidate RPs. The BSR selects RP-to-group mappings, which is called the RP set, and distributes the selected RPs using bootstrap messages sent to 224.0.0.13 (all PIM routers).

DVMRP

RFC 1075 describes DVMRP. It is the primary multicast routing protocol used in the multicast backbone (MBONE). The MBONE is used in the research community.

DVMRP operates in dense mode using RPF by having routers send a copy of a multicast packet out all paths. Routers that receive the multicast packets then send prune messages back to their upstream neighbor router to stop a data stream if no downstream receivers of the multicast group exist (either receiving routers or hosts on connected segments). DVMRP implements its own unicast routing protocol, similar to RIP, based on hop counts. DVMRP has a 32 hop-count limit. DVMRP does not scale suboptimally. Cisco's support of DVMRP is partial; DVMRP networks are usually implemented on UNIX machines running the mrouted process. A DVMRP tunnel is typically used to connect to the MBONE DVMRP network.

IPv6 Multicast Addresses

IPv6 retains the use and function of multicast addresses as a major address class. IPv6 prefix FF00::/8 is allocated for all IPv6 multicast addresses. IPv6 multicast addresses are described in RFC 2373. EIGRP for IPv6, OSPFv3, and RIPng routing protocols use multicast addresses to communicate between router neighbors.

The format of the IPv6 multicast address is described in Chapter 8, "Internet Protocol Version 6." The common multicast addresses are repeated in Table 12-3.

Table 12-3. Well-Known Multicast Addresses