CIR is a service guarantee from a frame relay provider of a certain end-to-end minimum bandwidth.
One benefit is that only packets sent above the CIR are marked discard eligible (DE). During times of network congestion, DE packets are the first to be discarded. All of your packets sent within CIR are "protected" from the first round of discarding. Also, by designating a CIR, you can determine which PVCs should be allocated the most bandwidth during times of bandwidth contention. You can use PVC CIR as a prioritization mechanism. Finally, your network design more closely matches your actual traffic patterns.
While CIR is an extremely useful concept and network metric, it can be quite difficult, if not impossible, to measure precisely.
The problem starts with the specification of exactly what a CIR is. In the ANSI T1S1 documents that originally specified frame relay, some key parameters were defined. These include B(c), B(e), and T(c), where "B" stands for "number of bits," "T" for "time, "c" for "committed" and "e" for "excess."
B(c) and B(e) are numbers of bits that may be transmitted within a specified time, T(c). Within T(c), the time over which the CIR is measured, the carrier guarantees that up to B(c) bits will be transmitted through the network under normal circumstances. Further, up to B(e) bits offered to the network within T(c) will be transmitted when network capacity is available. That is, these bits are "flying standby" and will have their discard eligible, or DE, bits set to indicate that they are of lower priority than B(c) bits.
Sounds pretty precise, right? But there's a problem. The time, or T(c), over which the bits are measured is not specified in the standards. T(c) is a component you have to negotiate with your carrier. By default, the T(c) is equal to the number of committed bits divided by the CIR, or T(c)=B(c)/CIR. This is nice, but meaningless, arithmetic. It's meaningless because this makes whether your CIR is specified in seconds, milliseconds or hours the determining factor for T(c). While a CIR of 32K bit/sec and 32 bits per millisecond are equivalent mathematically, in the former case you would measure CIR over a one-second interval, and in the latter you would measure it over a one-millisecond interval.
The bottom line is that having your carrier specify the interval over which the CIR is measured is a great starting point toward making the CIRs negotiated in your service leval agreement accurate and meaningful.
From CIR: Useful, but tough to measure, Network World on Frame Relay, 11/03/99.
Additional resources
Preserving CIR advantages
Network World on Wide Area Networking, 01/21/02.
Delivery guarantees without a CIR
Network World Frame Relay Newsletter, 04/03/00.
