The disclosure presents a system, method, and computer readable medium for computing network latencies. Network devices using flow processors create network traffic summary information in the form of flow records. A harvester may gather, process, and analyze flow records to find matching flows. Using matching flows, a harvester computes the latency, either one-way or round-trip, between two devices on the network. This information may be exported or stored on other devices on the network. The teachings of the present disclosure enable wide network coverage through the use of existing network traffic summary records. Further, the present disclosure mitigates network slowdowns associated with monitoring equipment. The present disclosure also reduces or eliminates the need for clock synchronization between devices on the network.