Reliable Queuing One-Way Delay Metric for Computer Networks

We claim the method to obtain a reliable queuing one-way delay metric for the wireless computer networking systems. The method requires only a measurement of the timestamp series between packet departure and arrival events. These timestamps are included inside the packets with stream frames. Thus, the method does not need to involve any additional probe packets nor any clock synchronization between two measurement nodes. We describe the algorithm based on the Gerchberg-Saxton procedure for queuing delay decoupling from a measured end-to-end delay distribution. We solve the problem of separating a queuing delay component from the fluctuating transmission and propagation components of delay. The reconstructed queuing delay distribution is further decomposed into individual spectral Weibullian components using the Expectation-Maximization algorithm. We propose an application of temporal queuing delay dynamics to infer a network congestion. Experimental results demonstrate the ability of early congestion detection before packet losses occur.