Using Multicast Domains
- Introduction to IP Multicast
- Enterprise Multicast in a Service Provider Environment
- mVPN Architecture
- Case Study of mVPN Operation in SuperCom
Multicast is a popular feature used mainly by IP-networks of Enterprise customers. Multicast allows the efficient distribution of information between a single multicast source and multiple receivers. An example of a multicast source in a corporate network would be a financial information server provided by a third-party company such as Bloomberg's or Reuters. The receivers would be individual PCs scattered around the network all receiving the same financial information from the server. The multicast feature allows a single stream of information to be transmitted from a source device, regardless of how many receivers are active for the information from that source device. The routers automatically replicate a single copy of the stream to each interface where multicast receivers can be reached. Therefore, multicast significantly reduces the amount of traffic required to distribute information to many interested parties.
This chapter describes in detail how an MPLS VPN service provider can provide multicast services between multiple sites of a customer VPN that has an existing multicast network or is intending to deploy the multicast feature within their network. This feature is known as multicast VPN (mVPN) and is available from Cisco IOS 12.2(13)T onward. This chapter includes an introduction to general IP Multicast concepts, an overall description of the mVPN feature and architecture, a detailed description of each IP Multicast component modified to support the mVPN feature, and a case study that shows how you can implement mVPN in an MPLS VPN backbone.
Introduction to IP Multicast
IP multicast is an efficient mechanism for transmitting data from a single source to many receivers in a network. The destination address of a multicast packet is always a multicast group address. This address comes from the IANA block 224.0.0.023188.8.131.52. (Before the concept of classless interdomain routing, or CIDR, existed, this range was referred to as the D-class.) A source transmits a multicast packet by using a multicast group address, while many receivers "listen" for traffic from that same group address.
Examples of applications that would use multicast are audio/video services such as IPTV, Windows Media Player, conferencing services such as NetMeeting or stock tickers, and financial information such as those that TIBCO and Reuters provide.
If you want to gain a more complete or detailed understanding of IP multicast, then read the Cisco Press book titled Developing IP Multicast Networks (ISBN 1-57870-077-9) or any other book that provides an overview of multicast technologies. You can obtain further information on advanced multicast topics from http://www.cisco.com/go/ipmulticast.
Multicast packets are forwarded through the network by using a multicast distribution tree. The network is responsible for replicating the same packet at each bifurcation point (the point at which the branches fork) in the tree. This means that only one copy of the packet travels over any particular link in the network, making multicast trees extremely efficient for distributing the same information to many receivers.
There are two types of distribution trees: source trees and shared trees.
A source tree is the simplest form of distribution tree. The source host of the multicast traffic is located at the root of the tree, and the receivers are located at the ends of the branches. Multicast traffic travels from the source host down the tree toward the receivers. The forwarding decision on which interface a multicast packet should be transmitted out is based on the multicast forwarding table. This table consists of a series of multicast state entries that are cached in the router. State entries for a source tree use the notation (S, G) pronounced S comma G. The letter S represents the IP address of the source, and G represents the group address.
The notion of direction is used for packets that are traveling along a distribution tree. When a packet travels from a source (or root) toward a receiver, it is deemed to be traveling down the tree. If a packet is traveling from the receiver toward the source (such as a control packet), it is deemed to be traveling up the tree.
A source tree is depicted in Figure 7-1. The host 184.108.40.206 at the root of the tree is transmitting multicast packets to the destination group 220.127.116.11, of which there are two interested receivers. The forwarding cache entry for this multicast stream is (18.104.22.168, 22.214.171.124).
A source tree implies that the route between the multicast source and receivers is the shortest available path; therefore, source trees are also referred to as shortest path trees (SPTs). A separate source tree exists for every source that is transmitting multicast packets, even if those sources are transmitting data to the same group. This means that there will be an (S, G) forwarding state entry for every active source in the network. Referring to our earlier example, if another source, such as 126.96.36.199, became active that was also transmitting to group 188.8.131.52, then an additional state entry (and a different SPT) would be created as (184.108.40.206, 220.127.116.11). Therefore, source trees or SPTs provide optimal routing at the cost of additional multicast state information in the network.
Figure 7-1 Source Distribution Tree
The important thing to remember about source trees is that the receiving end can only join the source tree if it has knowledge of the IP address of the source that is transmitting the group in which it is interested. In other words, to join a source tree, an explicit (S, G) join must be issued from the receiving end. (This explicit [S, G] join is issued by the last hop router, not the receiving host. The receiving host makes the last hop router aware that it wants to receive data from a particular group, and the last hop router figures out the rest.)
Shared trees differ from source trees in that the root of the tree is a common point somewhere in the network. This common point is referred to as the rendezvous point (RP). The RP is the point at which receivers join to learn of active sources. Multicast sources must transmit their traffic to the RP. When receivers join a multicast group on a shared tree, the root of the tree is always the RP, and multicast traffic is transmitted from the RP down toward the receivers. Therefore, the RP acts as a go-between for the sources and receivers. An RP can be the root for all multicast groups in the network, or different ranges of multicast groups can be associated with different RPs.
Multicast forwarding entries for a shared tree use the notation (*, G), which is pronounced star comma G. This is because all sources for a particular group share the same tree. (The multicast groups go to the same RP.) Therefore, the * or wildcard represents all sources. A shared tree is depicted in Figure 7-2. In this example, multicast traffic from the source host 18.104.22.168 and 22.214.171.124 travel to the RP and then down the tree toward the two receivers. There are two routing entries, one for each of the multicast groups that share the tree: (*, 126.96.36.199) and (*, 188.8.131.52). In a shared tree, if more sources become active for either of these two groups, there will still be only two routing entries due to the wildcard representing all sources for that group.
Figure 7-2 Shared Distribution Tree
Shared trees are not as optimal in their routing as source trees because all traffic from sources must travel to the RP and then follow the same (*, G) path to receivers. However, the amount of multicast routing state information required is less than that of a source tree. Therefore, there is a trade-off between optimal routing versus the amount of state information that must be kept.
Shared trees allow the receiving end to obtain data from a multicast group without having to know the IP address of the source. The only IP address that needs to be known is that of the RP. This can be configured statically on each router or learned dynamically by mechanisms such as Auto-RP or Bootstrap Router (BSR).
Shared trees can be categorized into two types: unidirectional and bidirectional. Unidirectional trees are essentially what has already been discussed; sources transmit to the RP, which then forwards the multicast traffic down the tree toward the receivers.
In a bidirectional shared tree, multicast traffic can travel up and down the tree to reach receivers. Bidirectional shared trees are useful in an any-to-any environment, where many sources and receivers are evenly distributed throughout the network. Figure 7-3 shows a bidirectional tree. Source 184.108.40.206 is transmitting to two receivers A and B for group 220.127.116.11. The multicast traffic from the source host is forwarded in both directions as follows:
Up the tree toward the root (RP). When the traffic arrives at the RP, it is then transmitted down the tree toward receiver A.
Down the tree toward receiver B. (It does not need to pass the RP.)
Bidirectional trees offer improved routing optimality over unidirectional shared trees by being able to forward data in both directions while retaining a minimum amount of state information. (Remember, state information refers to the amount of (S, G) or (*, G) entries that a router must hold.)
Figure 7-3 Bidirectional Shared Tree
Packet forwarding in a router can be divided into two types: unicast forwarding and multicast forwarding. The difference between unicast forwarding and multicast forwarding can be summarized as follows:
Unicast forwarding is concerned with where the packet is going.
Multicast forwarding is concerned with where the packet came from.
In unicast routing, the forwarding decision is based on the destination address of the packet. At each router along the path, you can derive the next-hop for the destination by finding the longest match entry for that destination in the unicast routing table. The unicast packet is then forwarded out the interface that is associated with the next-hop.
Forwarding of multicast packets cannot be done in the same manner because the destination is a multicast group address that you will most likely need to forward out multiple interfaces. Multicast group addresses do not appear in the unicast routing table; therefore, forwarding of multicast packets requires a different process. This process is called Reverse Path Forwarding (RPF), and it is the basis for forwarding multicast packets in most multicast routing protocols. In particular, RPF is used with Protocol Independent Multicast (PIM), which is the protocol used and described throughout this chapter.
Every multicast packet received on an interface at a router is subject to an RPF check. The RPF check determines whether the packet is forwarded or dropped and prevents looping of packets in the network. RPF operates like this:
When a multicast packet arrives at the router, the source address of that packet is checked to make sure that the incoming interface indeed leads back to the source. (In other words, it is on the reverse path.)
If the check passes, then the multicast packet is forwarded out the relevant interfaces (but not the RPF interface).
If the RPF check fails, the packet is discarded.
The interface used for the RPF check is referred to as the RPF interface. The way that this interface is determined depends on the multicast routing protocol that is in use. This chapter is concerned only with PIM, which is the most widely used protocol in Enterprise networks. PIM is discussed in the next section. PIM uses the information in the unicast routing table to determine the RPF interface. Figure 7-4 shows the process of an RPF check for a packet that arrives on the wrong interface. A multicast packet from the source 18.104.22.168 arrives on interface S0. A check of the multicast routing table shows that network 22.214.171.124 is reachable on interface S1, not S0; therefore, the RPF check fails and the packet is dropped.
Figure 7-4 RPF Check Fails
Figure 7-5 shows the RPF check for a multicast packet that arrives on the correct interface. The multicast packet with source arrives on interface S1, which matches the interface for this network in the unicast routing table. Therefore, the RPF check passes and the multicast packet is replicated to the interfaces in the outgoing interface list (called the olist) for the multicast group.
Figure 7-5 RPF Check Succeeds
If the RPF check has to refer to the unicast routing table for each arriving multicast packet, this will have a detrimental affect on router performance. Instead, the RPF interface is cached as part of the (S, G) or (*, G) multicast forwarding entry. When the multicast forwarding entry is created, the RPF interface is set to the interface that leads to the source network in the unicast routing table. If the unicast routing table changes, then the RPF interface is updated automatically to reflect the change.
Example 7-1 shows a multicast forwarding entry for (126.96.36.199, 188.8.131.52). You can also refer to this entry as a multicast routing table entry. The presence of the source in the (S, G) notation indicates that this entry is associated with a source tree or shortest path tree. The incoming interface is the RPF interface, which has been set to POS3/0. This setting matches the next-hop interface shown in the OSPF routing entry for the source 184.108.40.206. There are two interfaces in the outgoing olist: Serial4/0 and Serial4/2. The outgoing interface list provides the interfaces that the multicast packet should be replicated out. Therefore, packets that pass the RPF check from source 220.127.116.11 (they must come in on POS3/0) that are destined to group 18.104.22.168 are replicated out interface Serial4/0 and Serial4/2.
Example 7-1 Source Tree Multicast Forwarding Entry
(22.214.171.124, 126.96.36.199), 00:03:30/00:03:27, flags: sT Incoming interface: POS3/0, RPF nbr 188.8.131.52 Outgoing interface list: Serial4/0, Forward/Sparse-Dense, 00:03:30/00:02:55 Serial4/2, Forward/Sparse-Dense, 00:02:45/00:02:05 Routing entry for 184.108.40.206/32 Known via "ospf 1", distance 110, metric 2, type intra area Last update from 220.127.116.11 on POS3/0, 1w5d ago Routing Descriptor Blocks: * 18.104.22.168, from 22.214.171.124, 1w5d ago, via POS3/0 Route metric is 2, traffic share count is 1
For completeness, a shared tree routing entry is shown in Example 7-2. This entry represents all sources transmitting to group 126.96.36.199. The RPF interface is shown to be FastEthernet0/1, which is the next-hop interface to the RP 188.8.131.52. Remember that the root of a shared tree are always the RP; therefore, the RPF interface for a shared tree is the reverse path back to the RP.
Example 7-2 Shared Tree Multicast Forwarding Entry
(*, 184.108.40.206), 2w5d/00:00:00, RP 220.127.116.11, flags: SJCL Incoming interface: FastEthernet0/1, RPF nbr 192.168.2.34 Outgoing interface list: FastEthernet0/0, Forward/Sparse, 00:03:29/00:02:54
The outgoing interface lists in the preceding examples are determined by the particular multicast protocol in use.
Over the years, various multicast protocols have been developed, such as Distance Vector Multicast Routing Protocol (DVMRP), Multicast Open Shortest Path First (MOSPF), and Core Base Trees (CBT). The characteristic that these protocols have in common is that they create a multicast routing table based on their own discovery mechanisms. The RPF check does not use the information already available in the unicast routing table.
The protocol that is the most widely deployed and relevant to this chapter is PIM. As discussed previously, PIM uses the unicast routing table to discover whether the multicast packet has arrived on the correct interface. The RPF check is independent because it does not rely on a particular protocol; it bases its decisions on the contents of the unicast routing table.
Several PIM modes are available: dense mode (PIM DM), sparse mode (PIM SM), Bidirectional PIM (PIM Bi-Dir), and a recent addition known as Source Specific Multicast (SSM).
The deployment of PIM DM is diminishing because it has been proven to be inefficient in comparison to PIM SM. PIM DM is based on the assumption that for every subnet in the network, at least one receiver exists for every (S, G) multicast stream. Therefore, all multicast packets are pushed or flooded to every part of the network. Routers that do not want to receive the multicast traffic because they do not have a receiver for that (S, G) send a prune message back up the tree. Branches that do not have receivers are pruned off, the result being a source distribution tree with branches that have receivers. Periodically, the prune message times out, and multicast traffic begins to flood through the network again until another prune is received.
PIM SM is more efficient than PIM DM in that it does not use flooding to distribute traffic. PIM SM employs the pull model, in which traffic is distributed only where is it requested. Multicast traffic is distributed to a branch only if an explicit join message has been received for that multicast group. Initially, receivers in a PIM SM network join the shared tree (rooted at the RP). If the traffic on the shared tree reaches a certain bandwidth threshold, the last hop router (that is, the one to which the receiver is connected) can choose to join a shortest-path tree to the source. This puts the receiver on a more optimal path to the source.
PIM Bi-Dir creates a two-way forwarding tree, as shown in Figure 7-3. All multicast routing entries for bidirectional groups are on a (*, G) shared tree. Because traffic can travel in both directions, the amount of state information is kept to a minimum. Routing optimality is improved because traffic does not have to travel unnecessarily toward the RP. Source trees are never built for bidirectional multicast groups. Bidirectional trees in the service provider network are covered in the section "Case Study of mVPN Operation in SuperCom" later in this chapter.
SSM implies that the IP address of the source for a particular group is known before a join is issued. SSM in Cisco IOS is implemented in addition to PIM SM and co-exists with IP Multicast networks based on PIM SM. SSM always builds a source tree between the receivers and the source. The source is learned through an out-of-band mechanism. Because the source is known, an explicit (S, G) join can be issued for the source tree that obviates the need for shared trees and RPs. Because no RPs are required, optimal routing is assured; traffic travels the most direct path between source and receiver. SSM is a recent innovation in multicast networks and is recommended for new deployments, particularly in the service provider core for an mVPN environment. A practical deployment of SSM is discussed in the section, "Case Study of mVPN Operation in SuperCom" later in this chapter.
Multicast is a powerful feature that allows the efficient one-to-many distribution of information. Multicast uses the concept of distribution trees, where the source is the root of the tree and the receivers are at the leaves of the tree. The routers replicate packets at each branch of the tree, known as the bifurcation point. The tree is represented as a series of multicast state entries in each router, and packets are forwarded down this tree (toward the leaves) by using RPF. There are various modes of multicast operation in networks with the most popular one being PIM SM.