Section 4.5. Media Types
4.5. Media TypesThe preceding discussion on DRs and DISs makes it clear that both OSPF and IS-IS behave differently for different physical media. Understanding these differences is important not only for correctly designing your networks but also for troubleshooting problems on your network. 4.5.1. OSPF Network TypesOSPF classifies all interfaces, physical or logical, into one of five network types:
Broadcast networks might be more accurately named broadcast multi-access networks. More than two devices can connect to the network, and packets sent by one device can be seen by all connected devices. These days, broadcast networks almost always means Ethernet. Token Ring and FDDI, nowadays considered obsolete LAN technologies, are also broadcast networks. OSPF procedures assume bidirectional communication capabilities for all neighbors on broadcast networks, and hence that a single packet sent to one of the two OSPF multicast addresses AllSPFRouters (224.0.0.5) and AllDRouters (224.0.0.6) will be received by all neighbors on the network. A DR is always elected on a broadcast network. Point-to-point networks always connect just two neighbors. The physical topology directly corresponds to a branch on the SPF tree, so there is no need for a DR, and no DR is elected. If IP addresses are configured on the router's physical interfaces connecting to the link, all OSPF packets are sent to the AllSPFRouters multicast address. However, the very nature of a point-to-point network is that if a packet is transmitted onto the link, it has nowhere to go except to the router at the other end. So, OSPF will operate over unnumbered point-to-point linksthat is, point-to-point links with no IP addresses assigned. NBMA networks include Frame Relay and ATM and, if you are operating an ancient network, X.25. Common to all of these media is that the network is comprised of virtual circuits (VCs) connecting the attached devices. As with broadcast networks, OSPF assumes that NBMA networks are represented by a single IP subnet and that more than two routers can be connected to the subnet. The key difference is that there is no assumption that a packet sent by one router will be seen by all other routers on the subnet. If the network consists of a full mesh of VCs, as shown on the left of Figure 4.27, any router can send packets directly to all other routers on the network. But economy oftenif not usuallylimits the number of VCs to a partial mesh, as shown on the right. "Nonbroadcast" means that a single packet cannot be seen by all connected devices. Therefore, OSPF unicasts a separate packet to each of its known neighbors on NBMA networks. Figure 4.27. The virtual circuits comprising a NBMA network can be a full mesh or a partial mesh, so there is no guarantee that every connected device has a direct VC to every other device. Because an NBMA network is assumed to be a single IP subnet with possibly more than two routers attached, OSPF elects a DR and BDR. But because there is no assurance of a full VC mesh, the DR election process must be carefully managed. There are two choices:
Another factor to consider when working with a partially meshed NBMA network is that some interfaces, such as Frame Relay, do not by default support broadcast and multicast packets. As a result, OSPF cannot multicast Hellos onto the network to discover neighbors and cannot unicast Hellos without knowing who the neighbors are. There are, again, two solutions to this problem:
Figure 4.28 shows an example of the first option, using Cisco Systems IOS to perform the configuration. The interface configuration includes statements that map three neighbor IP addresses to the Frame Relay DLCIs that connect to the neighbors. Under the OSPF section, you can see that the addresses and the router priorities of the three neighbors are manually provided to OSPF. The priority of all three neighbors is 0, which means they are ineligible to become DRs. The router on which this configuration resides has a default priority, and the configurations on all of the neighbors referring to this router will assign a priority value greater than 0. As a result, this router becomes the DR.
Figure 4.29 shows the second option, in which broadcast (and by extension, multicast) support is added to the neighbor-to-DLCI mappings, allowing the OSPF network type to be changed to broadcast. Because Hellos can be exchanged using this configuration, router priorities are assigned only on the routers to which the priority applies. A priority of 50 is assigned to this router, and a priority of 0 is again configured for the neighbors so that this router will become the DR.
A simpler alternative to running OSPF over NBMA networks is to use the fourth OSPF network type, point-to-multipoint. OSPF point-to-multipoint treats the NBMA VCs as a collection of point-to-point links, and as a result no DR election takes place. Figure 4.30 shows an example using this network type. Notice that no router priority is assigned, because none is needed. There are also no static mappings of neighbor addresses to DLCIs; instead, Frame Relay uses inverse ARP to map dynamically the network to data link addresses.
The simplest approach to OSPF over NBMA networks springs directly from the way most VCs are configured in modern NBMA networks. Rather than treating the network as a single subnet attached to the routers' physical interfaces, multiple logical interfaces (also called subinterfaces) are configured on the physical interface, and each VC attaches to a logical interface. Each VC is then treated as a separate point-to-point link, either with its own subnet address or as an IP unnumbered link. With this underlying configuration, you simply specify the logical interfaces as OSPF point-to-point network types (see Figure 4.31).
The fifth OSPF network type, virtual link, cannot be easily described until OSPF area issues are discussed in more detail. Therefore, Chapter 7 describes virtual links. 4.5.2. IS-IS Network TypesIS-IS, in contrast to OSPF, supports only two network types (or subnetwork types, in IS-IS terminology):
Broadcast networks are the same in IS-IS as they are in OSPF: networks to which more than two devices can attach, and over which a single PDU can be received by all attached devices. As with OSPF, the primary characteristic of an IS-IS broadcast network type is that a DIS is elected to represent the network as a pseudonode. Also as with OSPF, there is an assumption that a single PDU sent to one of the two IS-IS multicast MAC addresses AllL1ISs (0180.c200.0014) or AllL2ISs (0180.c200.0015) can be received by all attached IS-IS routers. General topology networks are point-to-point links. The IS-IS spec breaks this network type down into several subtypes, but for our purposes it is sufficient to say that this second network type is point to point. IS-IS PDUs are unicast on general topology networks. ISO 10589 includes a specification for virtual links, but unlike OSPF IS-IS does not classify virtual links as a separate network type. The distinction is irrelevant for us anyway, because commercial implementations of IS-IS do not support the virtual links option. Chapter 7 says a bit more about IS-IS virtual links. The most noticeable difference from OSPF is that IS-IS has no equivalent of the OSPF NBMA and point-to-multipoint network types for NBMA support. But, as discussed in the preceding section, modern implementations of Frame Relay and ATM networks with very few exceptions treat each VC either as a distinct subnet or as an IP unnumbered link, rather than treating the entire NBMA "cloud" as a single subnet. As a result, both IS-IS and OSPF are normally configured to treat NBMA VCs as point-to-point network types. Figure 4.32 shows an example of an IS-IS configuration on a NBMA network. In this example, the network is ATM and the configuration is JUNOS rather than IOS. But the concept of configuring each VC as an individual subnet is the same as the example in Figure 4.31.
|
EAN: 2147483647
Pages: 111