The router provides a per-VC queuing architecture in the ATM-capable MDAs. In the egress direction toward the ATM port, the scheduling priority at the ATM layer is as follows:
CBR VCs are scheduled with strict priority over all other service categories.
rt-VBR VCs are scheduled next with strict priority over nrt-VBR and UBR VCs.
nrt-VBR shaped VCs are scheduled next with strict priority over nrt-VBR unshaped VCs and UBR VCs.
nrt-VBR unshaped VCs and UBR VCs are scheduled as a common class. Scheduling among these VCs is done using a WRR scheduler where the weight of each VC is determined by the configured SIR for nrt-VBR and by the MIR for UBR VCs. The scheduling is work-conserving, so each VC has access to excess bandwidth in proportion to its SIR/MIR. Under congestion, the performance of each VC degrades proportional to the weight of the VC.