Network Access, Resiliency, and Load Balancing

To ensure QoS, and for ease of configuration, an enterprise should prefer the use of "direct" access technologies between the CE and PE routers. That is, Cisco prefers the use of time-division multiplexing (TDM)-based (HDLC/PPP) access.

If indirect or shared access is used, such as Frame Relay or ATM, the enterprise needs to be assured that the proper QoS will be applied to support its jitter and committed bandwidth requirements. Frame Relay and ATM circuits must be provisioned such that the committed bandwidth rate is equal to the port rate specified on the PE routers. Any packets sent by a CE or PE over an Frame Relay or ATM-based access network must be guaranteed delivery by that access, because the CE and PE routers represent the primary IP QoS control points in the network. Because the Layer 2 parameters are defined this way, the drop and queuing decisions are maintained by devices with Layer 3 intelligence, and the Layer 2 connections are essentially straight pipes. For example, for Frame Relay, it is normally preferable for Layer 3 VPN deployments to configure the Frame Relay committed information rate (CIR) to be equal to the frame-relay mincir value in the map class. This means that no backward explicit congestion notifications (BECNs) should be generated and no drop decisions are made at Layer 2. Drop decisions are made by the application-aware IP mechanisms.

In addition, if ATM is used, the service provider should understand that the enterprise specifies link speeds in terms of IP bandwidth required. The ATM bandwidth required to carry that amount of IP bandwidth must account for the "cell tax," typically 40 percent over the IP bandwidth required. A single PVC should be used to deliver all IP traffic, conforming to jitter requirements.

For example, if the enterprise specifies that a site needs 2 Mbps of IP bandwidth, and the service provider can only offer ATM access, the service provider must provision a minimum of 2.8 Mbps of (sustained cell rate) bandwidth to support the IP bandwidth requirement.

In any case, where the PE port's speed exceeds the site's total committed bandwidth rate, which is likely in the case of Metro Ethernet, ATM, or FR, the PE routers must be capable of shaping all traffic to the committed bandwidth rate and providing the appropriate QoS into that shaped output queue.

In some cases, the enterprise provisions network access through two redundant access circuits between two different pairs of routers. The capability to load-balance traffic across these links for a larger aggregate capacity is required.

The following are examples of the types of questions to ask to obtain the provider capabilities just discussed:

• Access technologies What access technologies do you support? If access technologies differ based on location, please explain the differences between location types (such as tier 1, 2, and 3 sites) and what those mean in terms of access technology and speeds available. Should the access technologies differ significantly from the existing connections for your locations, the cost of new interface hardware may add significantly to the project's overall cost.

• Shaping capability In access technologies where the circuit speed is more than the committed bandwidth rate for a site, can you perform shaping for total output bandwidth? Can you also provide proper QoS handling for your classes of service into the shaped output queue?

• Load balancing In the case of redundant connections between two different PE routers and two different CE routers, does your network architecture permit equal-cost load balancing over them through the use of dual route distinguishers and eiBGP multipath?