Building Optical Control Planes: Challenges and Solutions

In many ways, the ASTN/ ASON model is quite traditional and incorporates aspects of other network technologies, such as the PSTN and ATM, although the current implementations for O-UNI and GMPLS are based on IP specifications.

Separate Planes
The control and transport planes of optical networks are quite separate, and therefore require the support of an out-of-band signaling solution, similar to the signaling system 7 (SS7) networks of the PSTN. In the case of optical cross-connect (OXC) transport planes, the optical control channels (OCCs) are connected to the transport plane by a connection control interface (CCI).

The dynamic aspects of the ASTN/ASON requirements (provisioning and restoration, for example) require complex interactions between the signaling and routing protocols, a mechanism equivalent to the UNI and PNNI interaction in ATM. The result is a dynamic, out-of-band control plane model where the signaling and data paths flow differently through the network.

Dynamic routing relies on direct information exchange between connected nodes, or neighbors, to discover topology information. To overcome the lack of direct connectivity required by a dynamic routing protocol, the out-of-band control plane is supplemented by a link layer node-to-node protocol.

The IETF has defined this link layer protocol as the link management protocol (LMP). This protocol is designed to supply the necessary information to the routing protocol, plus support traditional link layer operations, administration, and maintenance (OAM) functions, such as those performed by the SONET data communication channel (DCC).

Dynamic routing protocols, such as open shortest path first (OSPF) and intermediate system-to-intermediate system (IS-IS) use a simple metric to calculate the best path through a network. The ATM PNNI protocol also performs best path calculations, but includes support for QoS characteristics to determine the optimal path based on a number of additional criteria, such as bandwidth and delay.

Optical networks have their own QoS-like constraints, bandwidth being the most obvious, but also require additional parameters relating to such things as the characteristics of fibers, lambdas, latency, bundling, diversity, and jitter. OSPF and IS-IS use traffic-engineering extensions to propagate this QoS-related information. These traffic engineering (TE)-specific parameters are stored at each network node, and a modified constrained shortest path first (CSPF) engine is used to calculate a path through the network.

Once generated by a CSPF engine, this constrained route information is used by a signaling protocol, such as GMPLS, to establish the connection with each node, verifying that the resources are available as the signaling message transits the network. This check is necessary since a new connection may have been established and the network resources consumed, but the revised information may not have been flooded to every node in the network.

Additional Optical Complexities
There are additional complexities in the optical realm that also affect routing and can have a serious impact on control plane performance. For example, a single fiber connecting two optical switches may have many lambdas (see Figure 2). When a connection has been established at the lambda level, SONET/SDH connections can be established over the lambdas. If the assumption is made that a common set of protocols (such as those derived from MPLS/ GMPLS) can support both packet and optical/TDM services, then packets could flow over each of these connections.

Click here for Figure 2

Figure 2: Through GMPLS, designer can tap into a single set of protocols that support both packet and optical/TDM services. This greatly eases system and network design.

In effect, a hierarchy has been created with fibers at the bottom, followed by groups of lambdas (wavebands), individual lambdas, SONET/ SDH tributaries, and packet switch capable (PSC) connections at the top. As connections are established at each level of the hierarchy, they need to be propagated to other nodes in the network so they can potentially use these paths to establish higher-level connections. In the context of GMPLS, this can be accomplished by using forwarding adjacency label switched paths (FA-LSPs).

With the hierarchy and multiple connections between two OXCs, each connection could carry its own or multiple control plane channels. This could dramatically increase the control plane burden, with a severe impact on both cost and overall performance. But, by bundling lambdas together, a single control plane channel can support multiple physical connections.

With these enhancements in routing and signaling concepts, GMPLS and O-UNI can support all the requirements of optical signaling under the architecture laid out by the ASTN/ASON model. However, there are distinct differences between these protocols and their applications, which creates two possible network implementations: the peer GMPLS-only model and the O-UNI/GMPLS overlay model.

Comparing GMPLS, O-UNI
The interconnection of the IP router (client) and optical control planes can be defined as being either loosely or tightly coupled, where loose coupling refers to the overlay model and tight coupling refers to the peer model. There is also an augmented model, which is most closely related to the peer model, but shares some of the attributes of both models.

The peer model (GMPLS only) supports the I-NNI interface of the ASTN/ASON model and assumes that all devices in the network have a complete topological view, and that they participate more or less equally in routing. This would be supported, for example, in the collection of OCCs for a single network. In a full hierarchical network with optical switches, SONET/SDH add-drop multiplexers (ADMs) and IP routing devices, the entire optical core is visible to an edge router using the same interior gateway protocol (IGP) routing instance over the network, such as OSPF or IS-IS.

While this may provide a substantial opportunity for overall network optimization, it may not be an appropriate model to use if a carrier does not wish to expose critical OTN information (such as their network bandwidth, capacity and topology) to other carriers or divisions.

The GMPLS standard is essentially an extension of the classical MPLS protocols since it closely follows the "peer" network model of IP technology. The notable extensions in GMPLS that are distinct from MPLS, are the link-bundling and hierarchy capabilities described above, plus the bidirectional nature of optical connections, as opposed to the unidirectional LSPs that are established with classical MPLS.

The purpose of the overlay model (O-UNI/GMPLS) is to enable the client (the service requester) to add, modify, or delete connections to the carrier network, without providing the client any visibility to the carrier's network topology. The overlay model maintains a separation at the client-to-network interface by keeping the IP client routing, signaling protocols, topology distribution, and addressing scheme independent from the ones used by the optical layer of the carrier network.

Although the optical network may be allowed to use the same IP layer protocols as used in the client network for its routing and signaling, the client IP layer addressing must not dictate any addressing used in the carrier's optical control plane.

The augmented model, which is in the very early stages of specification, provides a mechanism for limited information sharing, typically using border gateway protocol version 4 (BGP-4) to pass the reachability information between optical networks. In the ASTN/ASON model, this represents the E-NNI interface and provides a full end-to-end multi-network implementation. This approach maintains the separation of the routing instances between the IP and optical domains, but passes the information from one routing instance to the other.

GMPLS/O-UNI Benefits

Having discussed the concepts behind the "intelligence" in the optical control plane, it is important to understand the benefits of using GMPLS and O-UNI vs. the adoption of some alternative protocol(s).

The GMPLS standards use traffic engineering (TE) extensions to the re-source reservation protocol (RSVP-TE) and similar extensions to constrained routing using the label distribution Protocol (CR-LDP). These augment traditional MPLS capabilities discussed above.

The OIF has further enhanced these standards to broaden the scope of the IETF support, particularly in the area of addressing. The IETF's MPLS is focused on IP technology, whereas GMPLS extends the addressing mechanism to nodes that may not be IP capable.

Clearly, some modification is required since, generally, it is assumed that every interface has an IP address and this is obviously a foreign concept with a DWDM connection. The OIF has extended the addressing capability to incorporate a wide range of options, including IPv4, IPv6, and the network service access point (NSAP) mechanisms.

With O-UNI and GMPLS using the same underlying signaling protocols and their close relationship to classical MPLS combined with the commonality of routing protocols, there are strong reasons to make them the signaling mechanism of choice for next generation optical networks.

Work on the ASTN/ASON, GMPLS, and O-UNI standards is progressing at a steady pace with some vendors already claiming to have field deployed pre-standard implementations. The challenges to full-scale adoption of these models lie in several areas: the remaining standards development work, persuading carriers to adopt IP technology in the control plane, and deriving the economic benefits carriers will accrue from deploying optical signaling technology.

The latter is perhaps the most challenging given the current economic climate. With reduced capital expenditures, service providers have shown increasing reluctance to invest in new technologies unless there is careful analysis of the increased revenue opportunity.

The ability to more rapidly provision bandwidth clearly enhances revenue by enabling service providers to support many new value-added services, and realize substantial cost savings. Optical signaling intelligence can be used to maximize the re-use of existing facilities, reducing the need to over- provision in order to ensure network capacity. This, combined with uniform signaling standards across multiple technologies such as DWDM and SONET/SDH, could translate into substantial savings in a relatively short time for both capital and operational expenditures.

About the Author

Deepak Shahane is the vice president and general manager of NetPlane Systems. He can be reached at deepak@netplane.com.