Predictive Lossless Queue for Extremely Large Dataset Transfer in Markov Communication Systems

Abstract

In this paper, a parametric prediction model is proposed for a queuing system driven by the Markov process. The aim of the model is to minimize the packet loss caused by time dependency characterized by the queue tail being too long, resulting in a time-out during the transfer of a large dataset. The proposed technique requires the key parameters influencing the queue content to be determined prior to its occupation regardless of the server capacity definition, using Bayesian inference. The subsequent time elapsing between the arrival and departure of a packet in the system is given, as well as the system load. This queue content planning is considered as the Markov birth-death chain; a type of discretization that characterizes almost all queuing systems, leading to an exponential queue, and captured herein using beta distribution. The inference results obtained using this exponential queue indicate that the queue with predictive parameters employing beta distribution, even when dealing with a loss system queue, involves less transition time and a greater load than a queue with coarse parameters; hence, preventing a long tail in the queue which is the cause of packet loss.