After defining services and their associated SLAs, the next step is to plan a roles-based network architecture. This network architecture specifies all the objects within a network, and their roles and relationships. It also defines technologies and the protocols that support them, as well as the topologies and transport mechanisms that connect the different objects.

Network Objects and Roles
A common, yet flexible, architectural approach for building service provider networks is to set up a four-layer architecture made up of a CPE layer, a physical aggregation layer, a service layer, and an intelligent core layer-all supported by end-to-end network management. Not every network needs to have each of these four layers-some networks may collapse multiple layers into one. This approach can be easily applied to service provider metro Ethernet networks, as well as to other technologies.

• The CPE layer - Usually managed by the service provider. This equipment is responsible for providing access security, and is also critical in enforcing SLAs for various applications. For example, CPE equipment might handle tasks like admission control, security policy enforcement, traffic classification, policing, and marking. Support for Layer 2 service creation (for example, ERS or EMS) also begins here.

• The aggregation layer - Cost-effectively aggregates traffic from the CPE layer and feeds it into the service layer devices. This layer supports local switching and reduces the amount of traffic passed along to the service layer.

• The service layer - Handles several sophisticated service functions, such as content distribution and enhanced security services like intrusion detection systems (IDSs). Firewall functions, hosted telephony, and Layer 2 service interworking are some of the other functions that this layer can handle. Ethernet service networks often choose to start MPLS core services at the service layer.

• The intelligent core - Efficiently transports traffic between service layer elements. It supports fast and reliable forwarding, along with sophisticated traffic engineering and congestion management. Although the intelligent core is not directly involved in individual customer networks, it can differentiate different traffic profiles and handle them accordingly.

• An extensive network management system (NMS) layer - Completes the network architecture, monitoring the network, and providing support for network configuration and troubleshooting. The NMS also supports end-to-end, point-and-click service provisioning and service monitoring, which helps to ensure that the SLAs defined in customer contracts are met.

Network Topology
The topology of how the different network layers are connected to each other and how the different elements within a layer are connected can play a major role for the overall scalability and cost of the network. The topology formed by the fiber that is already in the ground can have a major impact.

Rings or partial meshes are popular for two reasons. A ring can save providers fiber kilometers when compared to a hub-and-spoke architecture. If fiber is a scarce resource, rings are an excellent choice. A ring usually minimizes the amount of interfaces needed to interconnect several devices. In a ring topology, each device needs only two interfaces to connect to all the others.

Service providers should also consider that different topologies offer different strengths in terms of routing capabilities and protection protocols. Hub-and-spoke topologies are well suited for Layer 2 and Layer 3 routing protocols, whereas a ring is optimal for Resilient Packet Ring (RPR) or SDH/SONET protection.

Service providers should consider the convergence times of different protocols, and their effect on network availability. Spanning Tree Protocol is known to converge slowly, but 802.1w Rapid Spanning Tree can be a good alternative, especially in cost-sensitive environments. Providers should also think about the best ways to achieve redundancy at the network edge, and the convergence times required there.

By choosing the best interconnect medium, service providers can keep local traffic local to that portion of the network. This approach lets providers benefit from local switching, rather than have the traffic travel all the way to the core and back again (and create additional costs).

Scaling for Growth of Data Services
A major consideration for metro network architecture definition is scalability. Traditional "purpose-built" networks offered limited scalability, because they were designed to deliver only a specific service; any significant changes required expensive, manually-intensive equipment upgrades.

To illustrate the importance of scalability, consider what adding Ethernet services at a larger scale means to a traditional network-10,000 E1 (2-Mbps) services require the same switching capacity as 24 Gigabit Ethernet services. If a network was designed around 2-Mbps services and scaling steps, adding Gigabit Ethernet as a service introduces a major bandwidth challenge to this network.

With today's versatile transport and access solutions, however, service providers can choose from several technology options, using the scaling capabilities of Layers 1 to 3 to build scalable, flexible networks that support the services they have planned for. For example, they can add Layer 2 capabilities to Layer 1 equipment, enabling classical transport equipment to scale more effectively and support more services. To create a scalable, converged, multiservice network, service providers will probably use several transport options and protocols for service delivery.