• No results found

Basic Concepts From Computer Networks

CHAPTER 3 Frame Relay

3.1 Motivation and basic features

During the '80s, a number of significant changes took place in the environment in which communications networks operate. Transmission facilities were replaced by digital circuits based on optical fiber. The user equipment evolved from the dumb terminal to a powerful workstation which was capable of running large networking software and which could be directly attached to a LAN or a WAN. In parallel with these technological advances, there was an increasing demand for new bandwidth-thirsty applications, such

as moving large images and video conferencing. These new applications involved bursty traffic and required higher throughputs and faster response times.

Bringing all the above trends together, it became evident that in order to support the rapid transfer rates imposed by many new applications, and to provide the required response times, new high-speed WANs were needed. Due to the bursty nature of this traffic, these new networks should be based on packet switching rather than circuit switching. Frame relay and ATM networks are two such networks. Frame relay was originally defined to run over T1 and E1 links, and ATM networks were originally defined to run over OC-3 links, i.e., 155.52 Mbps. Both frame relay and ATM networks are connection-oriented packet switching networks. Frame relay was derived by removing and rearranging some of the functionality provided by some of the OSI protocol layers. ATM networks has many similarities to frame relay, but as will be seen in the next Chapter, they have a radically different architecture to frame relay and IP networks.

Older communication systems were designed based on the concept that the links between nodes were inherently unreliable. The ARQ mechanism in the data link layer was devised so that to guarantee an error-free transfer of packets over an unreliable link. The advent of fiber optics, however, made the data link layer ARQ mechanism redundant since fiber optics links introduce very few errors. Of course, there is no guarantee that the transmissions over a fiber optics link will be always error-free. Since it was not anticipated that there will be many retransmissions, it was felt that a considerable speed- up can be gained by removing the hop-by-hop ARQ scheme and simply rely on the end- devices, which had become quite intelligent, to recover packets that were either erroneously received or lost. Specifically, the recovery of these packets was left to the higher protocol layers, such as TCP, which run at the end-devices. This feature was implemented in both frame relay and ATM networks.

Another interesting development had to do with the speed of computers in relation to the speeds of the communication links in a WAN. In earlier computer networks, the

Physical Layer Network Layer Add packet envelop Add frame envelop Error control Rmve frame envelop Add frame envelop Error control Rmve frame envelop Rmve packet envelop Link Layer

Figure 3.1: Typical flow in X.25

Physical Layer Add envelop Rmve envelop Link Layer

Figure 3.2 Frame relay

communications links were very slow. As a result, a node, which was basically a computer system, could run the necessary networking protocols fast enough so that it did not delay the transmission of packets on the network links. In view of this, the network links were the bottleneck in the network. With the advent of fiber-based fast transmission links, the bottleneck shifted from the communication links to computers. That is, the software that executed the necessary protocol layers could not run fast enough to keep up with the new transmission speeds. In frame relay, significant software speed-ups were achieved by moving some of the functionality provided by layer 3 to layer 2. Specifically, the routing decision as to which node a packet should be forwarded to, which in X.25 and in IP networks is typically carried out in layer 3, was moved down to layer 2.

In figure 3.1, we show the path through the first three protocol layers that a packet follows as it gets switched through the X.25 network. This is contrasted in figure 3.2, where we show how switching is accomplished in frame relay. As can be seen in figure

3.1, the switching decision is made at the network layer and the error control and recovery is done at the link layer. The two layers introduce their own encapsulation overheads, and passing a packet from one layer to another requires moving it from one buffer of the computer’s memory to another, a time-consuming process.

In frame relay, as shown in figure 3.2, there is no ARQ at each hop and there is no processing at layer 3. Switching is carried out within the data link layer, thus avoiding additional overheads in memory transfers and encapsulation. A frame relay packet, known as a frame, is discarded by a frame relay node if its header is found in error. Also, a frame can be lost if it arrives at the input buffer of a frame relay node at a time when the buffer is full. We recall that in a network equipped with hop-by-hop ARQ, erroneously received packets or lost packets in a given hop, are recovered by the local

data link layer. In frame

LAN FR UNI FR UNI FR UNI FR net device FR net device FR net device FR net device Frame relay network Router Router Router LAN LAN

Figure 3.3: An example of the location of the frame relay UNIs

relay, discarded frames are only recovered by the end-user. As a result of these changes, frame relay provides a very efficient transport mechanisms.

As will be seen in the next Chapter, ATM networks have been built on similar principles.