Low overhead continuous monitoring of network performance

A method, apparatus, article of manufacture and computer product for low-overhead continuous monitoring of network performance in an intranet or Internet topology. Probe packets are sent from ingress access routers where they are received and processed by egress access routers. Probe packets are generated by copying every Nth packet being sent by an ingress access router. In the event an access router does not receive the probe packet, the probe packet is discarded through normal network delivery mechanisms. Network delay is determined by subtracting the time that a probe packet was received with the time stamp enclosed in the probe packet. Round trip time is established by reflecting the probe packet back to the originating access router and computing the round trip time. Bandwidth monitoring is achieved by using the number of probe packets received to estimate the expected amount of network traffic to be received. Fault monitoring is accomplished by comparing the number of probe packets received with the number of actual packets received. When the low overhead mechanisms indicate that network delays or faults exist, a heavy weight monitoring protocol is started between two access routers in question.