This paper describes an adaptive resource allocation approach and its Linux testbed implementation for QoS enabled Virtual Private Networks (VPNs). Specifically, we present and evaluate three schemes for predicting future traffic rates of a traffic flow from measurements. In our architecture, we also combine several QoS enabled network technologies. DiffServ over MPLS is used as the baseline in the pursuit for end-to-end QoS. Since a VPN operates by transporting traffic through tunnels, we use RSVP-TE to establish explicitly routed label switched paths (LSP). Dynamic Class Based Queuing (CBQ) was chosen to dynamically reallocate resources based on the prediction of future demands. Performance evaluations have been conducted, indicating superior performance.