Usually there are several PVCs operating on the access link with each VC having dedicated bandwidth availability. This is called the committed information rate (CIR). The CIR is the rate at which the service provider agrees to accept bits on the VC.

Individual CIRs are normally less than the port speed. However, the sum of the CIRs will normally be greater than the port speed. Sometimes this is a factor of 2 or 3. This statistical multiplexing accomodates the bursty nature of computer communications since channels are unlikely to be at their maximum data rate simultaneously.

While a frame is being transmitted, each bit will be sent at the port speed. For this reason, there must be a gap between frames on a VC if the average bit rate is to be the CIR.

The switch will accept frames from the DTE at rates in excess of the CIR. This effectively provides each channel with bandwidth on demand up to a maximum of the port speed. Some service providers impose a VC maximum that is less than the port speed. The difference between the CIR and the maximum, whether the maximum is port speed or lower, is called the Excess Information Rate (EIR).

The time interval over which the rates are calculated is called the committed time (T_c). The number of committed bits in T_c is the committed burst (B_c). The extra number of bits above the committed burst, up to the maximum speed of the access link, is the excess burst (B_e).

Although the switch accepts frames in excess of the CIR, each excess frame is marked at the switch by setting the Discard Eligibility (DE) bit in the address field.

The switch maintains a bit counter for each VC. An incoming frame is marked DE if it puts the counter over B_c. An incoming frame is discarded if it pushes the counter over B_c + B_e. At the end of each T_c seconds the counter is reduced by B_c. The counter may not be negative, so idle time cannot be saved up.

Frames arriving at a switch are queued or buffered prior to forwarding. As in any queuing system, it is possible that there will be an excessive buildup of frames at a switch. This causes delays. Delays lead to unnecessary retransmissions that occur when higher-level protocols receive no acknowledgment within a set time. In severe cases this can cause a serious drop in network throughput.

To avoid this problem, frame relay switches incorporate a policy of dropping frames from a queue to keep the queues short. Frames with their DE bit set will be dropped first.

When a switch sees its queue increasing, it tries to reduce the flow of frames to it. It does this by notifying DTEs of the problem by setting the Explicit Congestion Notification (ECN) bits in the frame address field.

The Forward ECN (FECN) bit is set on every frame that the switch receives on the congested link. The Backward ECN (BECN) bit is set on every frame that the switch places onto the congested link. DTEs receiving frames with the ECN bits set are expected to try to reduce the flow of frames until the congestion clears.

If the congestion occurs on an internal trunk, DTEs may receive notification even though they are not the cause of the congestion.

The DE, FECN and BECN bits are part of the address field in the LAPF frame.