Frame relay was designed to be a fast, reliable WAN service that could be used to extend the LAN services of a corporate environment. This is mostly true, but the problem with frame relay is that it cannot be used everywhere. XDSL is faster, typically just as reliable, and if available, can be used instead of frame relay. However, xDSL cannot be used everywhere either, which does not solve the initial availability problem created by frame relay or even ISDN. In addition, frame relay offers voice services on certain networks where DSL does not provide any service except data at this time. Since DSL is a broadband technology, which allows multiple digital signals to be used on the same bandwidth, in contrast to baseband, which only allows one, we are certain to see more services arise in the future.
More bandwidth is so badly needed that many WAN services were created in the 1980s and 1990s, with customers left on their own to solve the problem of availability for their remote locations. WAN standards were too slowly developed because companies were trying to own a certain technology instead of working together to create a large, fast WAN service. We can't blame them since they are in the business to make money, but no one WAN service has emerged as the bandwidth savior. In the future, we'll probably see more ATM networks since ATM is the fastest WAN service designed and implemented. Any new frame relay installations now run on top of ATM, if that gives you any indication of what the providers are planning.
The skinny on frame relay
Frame relay uses what are called permanent virtual circuits (PVCs) at the data-link layer of the OSI model. Depending on what Cisco documentation you are reading, physical layer specifications are also used in frame relay, but since it really doesn't matter what physical topology it runs on at the physical layer, nor what logical network addressing scheme you use at the network layer, typically you will discuss frame relay at the data-link layer only.
Believe it or not, frame relay was developed as a dial-up solution and ISDN was going to run on it at the network layer. That option is available from some providers (called switched virtual circuits or SVCs), but it is typically not used with frame relay networks.
The initial frame relay specifications described frame relay as a faster, more reliable packet-switching network than the existing X.25 packet-switching network, which had enormous overhead and very slow speeds (a whopping 19.2Kbps). Frame relay protocols assumed a better WAN network so reliability, like windowing and retransmission, was not needed. Interestingly enough, X.25 worked at the physical, data-link, and network layer protocols, which meant that logical addressing must be provided at the network layer by X.25. In the mid-1980s, X.25 was really a competitor to TCP/IP, which is likely another reason that frame relay replaced it, in addition to X.25's very slow speed.
X.25, ATM, and frame relay are considered a packet-switching network. The difference between a packet-switching network and a dedicated point-to-point network is that the available bandwidth in a packet-switching network is shared and the bandwidth is not shared in a point-to-point dedicated network.
Figure A shows the difference between a dedicated point-to-point network and a frame relay network.
Notice the three connections from the corporate office to the branches. In the dedicated point-to-point network, each line is dedicated to each branch and the bandwidth is available to that particular branch 24 hours a day, 7 days a week. (Well, as long as someone doesn't dig up the line, etc.) The downside to this is that the company must pay for this leased line even if it is not being used.
Since, typically, companies needed dedicated connections to each branch but transmitted bursty traffic instead of constant data traffic, frame relay became very popular, if available. However, each branch had to share the bandwidth with all the other branches.
Think of a packet-switching network as being similar to the telephone company's old party-line circuits. Party-line circuits were installed in neighborhoods where, to save money, less copper cable was pulled than needed, and houses in this neighborhood shared the same phone number.
This worked well in the 50s, 60s, and even into the 70s in some neighborhoods. However, this design was only successful if people were not on the phone all the time. As long as people only used the phone every once in a while, it was no problem. I remember when my older sisters were teenagers and my dad had to put in a dedicated line because the neighbors started complaining that they couldn't use the line. Packet-switching technology assumes the same type of usage and works well as long as companies are not trying to send constant data streams. If you need constant data streams sent to your remote locations, then frame relay is not for you.
Since frame relay had proprietary standards developed by each manufacturer, you had to install the same manufacturer's equipment from end to end if you wanted to run frame relay. Because of this, in 1984, the Consultative Committee on International Telephone and Telegraph (CCITT) tried to develop some standards for frame relay that would allow all (or most) vendors to interoperate. It didn't work out, as no one developed these standards at this time.
However, in 1990, Cisco led a group called the "group of four" to focus on frame relay technology. The four companies were Cisco, Digital, Northern Telecom, and StrataCom. They used what the CCITT started with and extended the protocol with features and additional capabilities for more complex networks. These features are known as the Local Management Interface (LMI) extensions. After the group of four's success, the CCITT as well as ANSI developed their own LMI types, which means that when you configure frame relay, you have to configure the router with the correct LMI extension type. Cisco's routers all use their own LMI extensions by default, naturally. So, if you are configuring frame relay between a Cisco and 3Com router, for example, you would configure the ANSI LMI type on both routers so they can communicate with the correct extensions.
Data-link layer addressing
At the data-link layer, frame relay uses Data Link Connection Identifiers (DLCIs) to identify each PVC. If you are sending a packet to a remote node, the logical network address must be resolved to the DLCI number of the PVC used to reach the remote network where the node is located. Frame relay provides this service dynamically using Inverse Address Resolution Protocol (IARP).
This does not work much differently from how a host on a LAN sends packets to a default gateway. Think of the PVC number as the default gateway address. When a host needs to send a packet to a remote network, the IARP looks up the DLCI number of the PVC needed to transmit the packet to the correct network, just as ARP on a LAN would find the hardware address of the configured default gateway.
Frame relay is a great WAN protocol that can help save companies money because they no longer have to pay for dedicated point-to-point connections. Companies using frame relay can share the bandwidth, and the cost, with another company as long as they both only need to send bursty data.