On the surface, Frame Relay is a marvelous technology for lowering ones' leased line network costs. However, there are traps for the unwary hiding in the very structure of the service.
The theory is nice:
everyone buys an "access pipe" into the frame relay switching fabric (e.g. a T1)
Permanent Virtual Circuits (PVC) are set up by the TelCo at customer direction from access point A to access point B through the frame relay cloud, each with a specific Committed Information Rate (CIR) (bandwidth, e.g. 64Kb/s, or 256Kb/s). With these PVCs, a network manager can design whatever kind of network he likes (e.g. star, mesh, amorphous, etc).
The access pipe is always faster than the sum of all the CIRs of all the PVCs terminating at your access.
The most attractive thing about Frame Relay is that if one or more of your PVCs needs to "burst" faster than the CIR (i.e. consume more bandwidth than the CIR allows), well, then it can do so, provided that there is available bandwidth in the network (from your access pipe on out). Nothing additional is required - you just start blatting more data through the PVCs. If the bandwidth is available in the network, your extra frames will arrive at their destination, and if not, they will be dropped by the Frame Relay switches.
The basic assumption in this network design is that no one will be using their full CIR on every PVC all the time (or even at the same time), so that there should always be "excess" bandwidth available for "bursting" above the CIR on any given PVC.
Unfortunately, this design also has some potentially nasty side effects given the economic structure of the offering:
The Customer has incentive to under-specify his bandwidth needs, because he can "burst" above the CIR to use the "idle" or "excess" bandwidth that other customers are not using at any given moment. However, if all customers behave this way...
The Provider has incentive to engineer the network with no excess or slop at all to keep his profit margins up - after all, the customer is only paying for his CIR, except for growth, why have any excess in the network at all? However, if a significant fraction of the customers have under-specified their CIRs...
The end result of this clash of incentives is a network that cannot burst above CIR on any PVC. There are only two forces that mitigate against this dire outcome:
Customer mistrust of the TelCo might lead them to specify a CIR in excess of their actual needs for any given PVC, or even all of their PVCs.
TelCo requirements for network growth to handle new business requires them to have excess bandwidth in the network at any given time, assuming that they wish to expand.
Telephone companies, whose principle business is voice, have consistently failed to provide data services with the same reliability, care and regard as their voice networks. Anecdotal evidence suggests that for any "intelligent" switched data service (e.g. Frame Relay, SMDS, ATM), whenever the switch configuration is touched by the TelCo (e.g. to add or delete a customer) one or more existing customers are adversely affected (i.e. lose service).
As is often said, "They Just Don't ``Get It.''"
The one data service that the TelCos seem to be able to get right with some consistency is leased circuits with no switching on them at all.