A method of dealing equitably with traffic from multiple VPNs feeding into
a single router utilizes the weighted fair queueing (WFQ) technique of
the prior art in combination with a "service level" queue comprising a
set of tokens defining the class of service (COS) with the next packet to
be transmitted and a queue selector that functions to query each queue in
a round robin fashion. The service level queue ensures that the higher
priority traffic will be transmitted first, while the queue selector
ensures that each VPN will receive similar service. The queue selector
may be "weighted" in that if a particular VPN generates substantially
more traffic than the other VPNS, that VPN will be selected more than
once during a round robin cycle to transmit a packet ("weighted round
robin" WFQ).