6.2. Types of Adjacencies and Discovery Procedures

ITU-T recommendation G.7714 [ITU-T02a] defines three types of adjacencies. These are described below.

6.2.1. Layer Adjacency

Consider Figure 6-1a, which illustrates a network segment with IP routers, OEO (e.g., SONET) optical cross-connects (OXCs), WDM devices, and amplifiers. The figure depicts physical links between network elements, for example, between the OXCs and the WDM devices. The transmission system, consisting of the WDM devices and optical amplifiers, is referred to as the optical line system. Figure 6-1b depicts connectivity at various SONET layers—path, line, and section (see also Figure 5-5). Two network elements are said to be neighbors at a certain layer if each terminates the overhead information for that layer generated by the other, and all the intermediate network elements pass this information transparently. For example, the two OXCs shown in Figure 6-1a are neighbors at the line layer. The connectivity at each layer, as shown in Figure 6-1b, defines logical links at that layer as seen from the layer above. Thus, layer adjacency defines the associations between the two end points that terminate a logical link at a given layer. In reality, the association is defined by a mapping of the two end point identifiers. For instance, if the end points are identified by two numbers i and j, respectively, the adjacency is defined by recognizing that end point i on one side is related to end point j on the other side. This recognition is brought about by running the neighbor discovery procedures, and the resulting mapping between the end point identifiers is used in signaling and routing (see, e.g., Chapter 7).

In Figure 6-1a, there are six links between the two OXCs, all carried over the same fiber. Each of these links could in fact support multiple communication channels.^[1] For instance, suppose each link is a channelized OC-48 link, capable of carrying 48 STS-1 connections. Each channel has to be recognized by an identifier at each end of the link. These identifiers, however, can be the same on both sides, since the channels have a fixed association to the link. For example, the first channel can be identified as channel 1, the second as channel 2, and so on, on each side. Thus, there is no need for an explicit discovery mechanism to detect channel associations within a logical link.

^[1] In the terminology of Chapter 5, such a logical link is in fact a server layer trail supporting multiple client layer link connections. In the interest of simplicity, we choose to be informal in this section.

Let us now look at some other configurations and see how layer adjacency is defined. Figure 6-2a depicts two SONET STS-1 cross-connects interconnected through an intervening core network supporting only STS-48 services. Each STS-1 OXC has an OC-48 interface to the core network, and the network supports line-transparent STS-48 connections (see Chapter 3). Figure 6-2b shows such a connection between the two STS-1 OXCs. This connection, from the point of view of the two OXCs, looks like a channelized STS-48 link. Thus, there is a line layer adjacency between the two OXCs. Similarly, there is a section layer adjacency between each STS-1 OXC and an OXC in the core network. As before, there is a fixed association between the STS-1 channels and the STS-48 link that runs between the two STS-1 OXCs.

Figure 6-3 depicts the connectivity at various layers of the network segment shown in Figure 6-1a, when Photonic Cross-Connects (PXCs) are used. A PXC cross-connects the optical signal from an incoming port to an outgoing port without converting it to the corresponding electrical signal. Hence, a PXC performs OOO switching (see Chapter 1), and it is transparent to the format of the signal. Despite the transparency, the connectivity between the PXCs still defines the optical network topology from the perspective of provisioning and restoration of path-layer connections. But at what layer are the PXCs adjacent? Figure 6-4 shows the anatomy of a PXC. It is seen that the PXC interfaces are connected to WDM transponders that determine the rate and other characteristics of the link between two PXCs. For instance, this could be an OC-192 link, although none of the SONET overheads are terminated by the PXCs. In this scenario, it might be appropriate to say that the PXCs are adjacent at the physical layer. In reality, the precise classification of this adjacency does not matter as much as the identification of the two end points of the link.

6.2.2. Physical Media Adjacency

Physical media adjacency, as the name implies, describes the adjacency between two network elements that have a direct physical connection. For example, considering Figure 6-1a, the OXCs have direct connections to WDM devices, that is, each OXC port is connected to a port in a WDM device. The association between the identities of the interconnected ports defines the physical adjacency.

6.2.3. Control Adjacency

In a network that utilizes control communication between NEs, it is necessary for each NE to know all its control plane adjacencies. For instance, OXCs 1 and 2 in Figure 6-1a could be peers in the control plane if signaling were to be used for connection provisioning (see Chapter 7). An association of the higher layer protocol identifiers for the two control end points defines the control adjacency. For example, suppose the GMPLS RSVP-TE protocol (see Chapter 7) is used for signaling between the two OXCs. The control channel between the two NEs could be identified by the interface IP address at each end. Neighbor discovery procedures can aid in automatically discovering control plane adjacencies.

6.2.4. Neighbor Discovery Procedures

Now that we have defined three types of adjacencies, the questions are which of these types of discovery are needed in a given network and how is discovery done? To answer these questions, we have to look at the role of discovery in some detail. We briefly mentioned earlier that discovery is essential for route computation and signaling. Let us see why this is so.

Figure 6-5 depicts a scenario for provisioning connections in an optical network via a centralized provisioning system. Here, to provision a connection from port 1 in node A to port 2 in node D, it is necessary to determine the precise path in terms of all the intermediate nodes and ports. This implies that the central provisioning system must have a detailed representation of the network topology, including information about

• The identities of OXCs in the network

• The manner in which OXCs are connected, that is, the identities of nodes and interfaces at the two ends of each link

• The type of each link, for example, channelized OC-48

• The properties of each link as pertaining to connection routing. These properties include

  • Shared-risk link group information: An SRLG is a unique identifier associated with physical facilities (e.g., a fiber or a conduit) and denotes a common or “shared” level of risk with other members of its group. Each link can be associated with one or more SRLGs depending on the physical facilities it utilizes. The assignment of SRLGs to links allows the computation of physically diverse paths.

  • Link span distance (also called fiber miles): This information indicates the geographic distance (e.g., number of miles) spanned by the link.

  • Link cost (also called Traffic Engineering (TE) metric): A number that indicates the administrative cost assigned to the link.

  • Administrative color: This is usually a number that indicates an administrative grouping for the link. Links with similar color values are said to be in the same administrative class with respect to applicable policies.

Thus, it is necessary to gather many different items of information in the provisioning system. Furthermore, the information must be updated as changes occur in the network. Automated procedures that aid in this process by minimizing the amount of manual configuration would be quite useful. Suppose that each OXC could automatically discover the identity of the remote OXC and the identity of the remote interface to which each of its local interfaces is connected. Furthermore, suppose that any changes in this information can be detected automatically and quickly. Then the provisioning system can simply collect up-to-date connectivity information from the individual OXCs to form the network topology map. This procedure, for a two-node network, is shown in Figure 6-6. Here, each node periodically sends over each link its own identity and the identity of the interface. Each node also keeps track of the information received from its neighbor. The provisioning system collects the information from both nodes, performs the correlation, and obtains the accurate connectivity information.

The alternative to this procedure is to manually enter the connectivity information in the provisioning system. This could be a rather cumbersome task, considering the number of ports in new generation optical switches and the need to keep track of connectivity changes as they occur.

A side benefit of automatic link identification is the ability of OXCs to detect inconsistent wiring of bidirectional interfaces. Specifically, it is possible to automatically detect situations where the transmit side of interface x in OXC A is connected to the receive side of interface y in OXC B, but the receive side of interface x is not connected to the transmit side of interface y. This feature is very useful when OXCs with high port densities are interconnected. This is further described later in this chapter.

Instead of the centralized provisioning system shown in Figure 6-5, let us now consider the case where distributed IP-centric routing and signaling protocols are implemented in the network. Here, provisioning activities are decentralized, and each OXC must keep track of the topology and the link information that a centralized provisioning system would have normally maintained. Note that in the distributed system configuration, the topology information maintained by a node need not be as detailed as that maintained by a centralized provisioning system. Specifically, information about links between OXC may be aggregated or bundled so that connectivity information may be abstracted. This concept is illustrated in Figure 6-7, where nineteen links between a pair of OXCs are abstracted into four bundles based on link type and SRLG values. So, with link bundling, is there a need for detailed link identification as in the centralized provisioning case? Yes! This information is now needed during the signaling phase of connection provisioning. Figure 6-8 illustrates the provisioning of a connection under distributed control between an ingress and an egress port (i.e., port 1 of OXC A to port 2 of OXC D). The source OXC (A) computes an explicit route (ER) which specifies the sequence of OXCs from the ingress to the egress, along with an identification of the specific link bundle at each OXC. In this example, there is a single bundle between each node pair, denoted by B1. Each intermediate OXC is responsible for selecting a specific link within the bundle to route the connection, and signaling the choice to the next OXC in the explicit route. The figure depicts the explicit route received by each intermediate OXC and the link selected by it. Since the link identifier could be different in the two ends, neighboring OXCs must agree on the identification of each topological link prior to signaling. The procedure depicted in Figure 6-6 could be used for this purpose. This would allow an OXC to know the identity of a link on the remote side and indicate it in signaling messages.

Those familiar with IP routing may wonder why the link identification procedure shown in Figure 6-6 is required if a distributed IP routing protocol (e.g., OSPF) with optical network extensions is running in each OXC. Such a protocol already incorporates an automatic adjacency discovery function. The reason for this is that even though an IP routing protocol might run in each OXC, there is no need to have a routing adjacency over each link. Indeed, as described in Chapter 10, the routing and signaling adjacencies in densely connected optical networks should be independent of the number of links between neighboring OXCs. To this end, it is desirable to create a separate procedure for the discovery of adjacencies and identification of links.

Thus, layer adjacency discovery is performed primarily over topological links, that is, links of significance to route computation. The specific layer in which adjacencies are determined depends on the scope of the routing procedures. For instance, in the network of Figure 6-1a, the line layer adjacency between the OXCs is important for routing SONET connections within the optical network. Outside of this network, the path layer adjacencies between the IP routers (which define the IP layer links) needs to be determined to aid in IP routing. Likewise, in the example of Figure 6-2, the line layer adjacencies between the STS-1 OXCs need to be discovered to route STS-1 connections in the outer network. Within the core network, however, the adjacencies between the STS-48 OXCs need to be determined.

In all these examples, physical adjacency was not a concern. It may be useful, however, to determine this for diagnostic and network management functions. As for control adjacency discovery, it is relevant only when control channels for signaling and routing need to be established between network elements. In other words, this is necessary only with a distributed control plane.

In the rest of this chapter, we focus on layer and control adjacency discovery. In this regard, the discussion that follows answers the question about how discovery is done. At a high level, discovering a layer adjacency requires some communication over the corresponding logical link. The content and format of this communication depends on the means available for such communication, as described next.