Design Parameters
| The following parameters must be considered in the Frame Relay design stage:
Bc, Be, Tc, and CIR are defined per data-link connection identifier (DLCI). Because of this, the leaky-bucket filter controls the rate per DLCI. The access rate is valid per User-Network Interface (UNI). Bc, Be, and CIR incoming and outgoing values can be distinguished. If the connection is symmetrical, the values in both directions are the same. For PVCs, incoming and outgoing Bc, Be, and CIR are defined at subscription time as follows:
If the Tc is one second:
CIR OptionsFigure 15-1 provides a general and simplistic view of CIR.[1] The CIR is computed over the minimum increment of Tc. Therefore, if Tc = 1.125s, a CIR of 64 kbps permits a Bc of 72 kbps (72 / 1.125 = 64). If the Bc is 144 kbps and the Tc is 1.125s, the CIR is 128 kbps (128 x 1.125 = 144). Figure 15-1. Committed Information Rate (CIR), Excess Information Rate (EIR), and the Service Offering
NOTE Each user's frame is sent at the wire (link) speed (access rate). The end user can be provisioned for a CIR of 128 kbps on a T1 link of 1.544 Mbps, and as a result, the traffic can burst across the link at 1.544 Mbps, which provides low latency in the ingress and egress points. Several factors must be considered when selecting a CIR from the service provider. Some service providers offer CIRs in increments of 16 kbps, from 0 up to a fractional T1, full T1, and more. Others might not offer CIR = 0 (zero CIR) and might provide CIRs in increments of 4 kbps or 64 kbps. Another choice can be between 56 kbps and fractions of a T1. The cost of the service is proportional to the CIR; therefore, the choice can significantly impact the monthly network charges for the enterprise. Zero CIR SelectionIf the design of the network is to carry only non-delay critical data, you should provision the circuit for CIR = 0. If the CIR = 0, every bit is marked with DE = 1. The fact that every frame is eligible for discard does not mean that the network drops the frame. In fact, service providers generally allow more than 99 percent of frames to flow from end to end. If the frame is dropped, the higher-layer protocols can retransmit the dropped packets. However, the frames can be delayed slightly until the network congestion clears. This is a trade-off for the lower cost of the service. Also, if all the endpoints are connected to the same Frame Relay switch, only the backplane of the switch is between them, and a CIR of 0 is often all that is needed for wire speed performance. The importance of CIR increases when the topology is more complex, such as in interswitch links. During times of network congestion, DE packets are the first to be discarded, but the data won't be lost because the upper-layer protocols will retransmit them. Aggregate CIRUsing an aggregate CIR, also referred to as oversubscription, can be a beneficial technique when traffic bursts occur on different PVCs. For example, if a T1 is installed to a core router, and four users are each configured for four slots of 64 kbps (64 x 4 = 384 kbps), two basic line conditions exist in the PVC design: idle and busy. If only two users use the service and the other two are idle, the first two can use the entire rate up to 1.544 Mbps, and because every user is setup for 384 kbps, every extra frame is marked with DE = 1. Regardless of the configured rate, the overall rate is much higher. Now, suppose that every user is configured for 768 kbps. The aggregated rate of all users configured is 2 x 1.544, which is referred to as an oversubscription. In this example, the ratio is 2:1, where the subscription rate is two times more (2 x 1.544) than the T1 speed (1.544). NOTE Many people are uneasy with the idea of oversubscription, sometimes referred to as statistical multiplexing. Telephone companies (telcos) get a bad reputation for doing it, but Internet service providers (ISPs) do it to a much greater extent. Oversubscription of 7:1 or more from an ISP is not uncommon. Usually, a 3:1 ratio is a sound ratio, depending on usage patterns. The important factor is to have a relatively reliable picture of the traffic distribution during the day, and the estimation that not all users will burst at the same time of day. Then, the 3:1 ratio reflects a traffic prediction that no more than 33 percent of users will use the service during the same period each day. Asymmetrical CIRsFor interactive query-response types of traffic and for typical client/server operations, the traffic in both directions can differ significantly. For this reason, it makes sense to consider different CIRs in both directions. This is referred to as asymmetrical CIRs. They reduce the organization's service charges if they are appropriate for that organization's traffic. Adjustable (Flexible) CIRIf the traffic can be planned precisely (for example, trending of the production environment with different shifts and number of users), and if the service provider supports this feature (for an additional fee), it is worth it to request an adjustable CIR. This feature is also known as bandwidth on demand and it can be beneficial when you are trying to satisfy expected changes in bandwidth requirements. When requesting adjustable CIRs from the service provider, it is important to keep in mind the following parameters:
UNIThe Frame Relay standards (refer to FRF 1.2) clearly define bearer services provided by the network to make UNI possible. As shown in Figure 15-2, the UNI operation area is locally defined between the data terminal equipment/data communications equipment (DTE/DCE) peers. In turn, the bearer service definition includes the following:
Figure 15-2. Area of Operation for the User-Network Interface (UNI)
The underlying assumption of Frame Relay is low error rates. Therefore, only a simple, connection-oriented transport service is required. The flow-control and error-recovery functions are beyond the scope of Frame Relay and are provided by higher-layer protocols, such as the Transmission Control Protocol (TCP). Therefore, the reduced overhead of Frame Relay in comparison to other similar technologies is one of its greatest advantages, providing high-rate and large payload connections. NOTE The opposite of X.25, which was the transport of choice prior to Frame Relay, TCP's ability to detect errors and the robustness of current carrier quality networks pushed the role of flow control and error recovery from Layer 2 to Layer 3/4. NNIFRF2.1 defines the NNI, which extends user communications beyond the scope of UNI. The NNI interface is concerned with the transfer of C-plane and U-plane information between two network nodes that belong to two different Frame Relay networks. Figure 15-3 shows the area of operation of NNI. To keep the architecture functioning in a multivendor environment, network-to-network signaling must exist to keep the PVCs interconnected. To provide this functionality, every pair of adjacent networks (1 and 2; 2 and 3) must support additional provisions to request status information (see details in Chapter 16), and must be able to respond to them. These provisions are called bidirectional procedures. More information about this topic can be found in the Frame Relay Forum's (FRF's) "Frame Relay Network-to-Network Interface Implementation Agreement," or at www.frforum.com. Figure 15-3. Area of Operation for the Network-to-Network Interface (NNI)
Voice over Frame RelayVoice over Frame Relay (VoFR) is defined under FRF.11.1. The Frame Relay approach requires the upper-layer protocols, such as the connection-oriented protocol TCP, to ensure end-to-end delivery of frames. However, retransmission results in repeating part of the voice message, which is not acceptable from a user's standpoint. In this case, packetizing and buffering of the frames is required, where the entire message must be stored and played with the appropriate timing. Imminent transport layer retransmissions are not the only possible issues in VoFR designs. Other issues are related to voice compression, silence suppression, echo cancellation, and the ability to preserve voice quality. Additional information and the latest developments on VoFR are available from the FRF, and at www.cisco.com. Frame Relay MulticastThe multicast provision exists in most local-area network (LAN) technologies and as part of the Internet Protocol (IP). Multicast is a feature that enables one source to send information to multiple recipients. A typical wide-area network (WAN) uses a point-to-point (unicast) connection, where the user transmits information to only one recipient. Frame Relay multicast is addressed in FRF.7, and this agreement defines one-to-many types of connection, where one sender who provides information to multiple recipients. The architecture is based on the International Telecommunication Union Telecommunication Standardization Sector (ITU-T) recommendation X.6, "Multicast Service Definition." The model defines multicast groups, which includes a multicast server that provides service to all members of the group. In the case of Frame Relay, the server must be internal to the network, so that members can establish multipoint relationships and participate in point-to-multipoint data transfer. Three types of services are defined:
In one-way service, a central location called a root has a point-to-point relationship with all other members of the group, which are called leaves. The root has a one-to-one connection to the server, as does every single leaf. Two-way multicast doubles the previously described structure, providing a full duplex connection. In N-way service, all members of the group are peers, thus every message is sent to all members of the group. The LMI extension provides the following multicast messages to ensure proper functioning of multicast groups:
|
EAN: 2147483647
Pages: 235