Recent investigations of internet of things network traffic suggest the failure of traditionally used Poisson process-based models. The paper substantiates this finding by performing statistical analysis of packets inter-arrival time in publicly available datasets of IoT-based sensor data traffic. It is observed that gamma distribution fits well to the inter-arrival time of packets. Motivated by this outcome, a detailed performance evaluation of Gamma/M/1 model is carried out. The closed form expressions of mean queue length, mean waiting time in the queue and overflow probability are derived. Numerical computations show that the mean queue length of Gamma/M/1 model explodes much faster than Poisson process based M/M/1 model when its shape parameter is low and scale parameter is high, whereas, the opposite is observed when the values of parameters are reversed. Monte-Carlo simulation of the Gamma/M/1 model also validates these findings. These results will be useful in dimensioning resources like buffers and in designing congestion-control algorithms for IoT systems.
The presence of diverse traffic types is well-established in IoT networks. Various probability distributions have been found to describe packets inter-arrival time contrasting with the familiar exponential distribution in traditional networks. These findings suggest the need to develop appropriate traffic models for performance analysis of IoT network systems. An essential component in IoT network is gateway as it provides connectivity to the core Internet. The IoT gateway also performs functions such as protocol translation and traffic aggregation. Therefore, efficient design of the IoT gateway is necessary for better network management. The paper presents a new analytical model, N-Gamma/M/1, for analyzing the performance of IoT gateway. The equivalence of N-Gamma/M/1 and Gamma/M/1 models is proved mathematically. Additionally, an in-depth performance evaluation of the IoT gateway under various arrival patterns is conducted through simulation. The numerical analysis of the proposed N-Gamma/M/1 model emphasizes the need for more buffers at the gateway when traffic from input devices has varying values of gamma distribution parameters. It is also noted that the IoT gateway experiences a longer mean queue length resulting in higher mean waiting time and packet loss when inter-arrival time distribution of packets follows generalized Pareto, Weibull and lognormal distributions with different parameter values. This makes the task of IoT network management challenging. Adaptive and intelligent resource allocation policies along with dynamic congestion control algorithms may provide a solution to minimize packet loss and ensure quality of service.
Statistical analysis of IoT network traffic establishes the presence of lognormal distributed inter-arrivals. The paper presents analysis of LN/M/1 queueing model using approximate results of Laplace-Stieltjes transformation of lognormal distribution in terms of Lambert W(.) function. The comparative performance evaluation of the LN/M/1 model with traditional M/M/1 model reveals that the quality of service metrics of LN/M/1 model mostly remain lower than M/M/1 implying that less number of buffers are required in IoT systems driven by such traffic. The asymptotic expressions for loss probability are derived using finite capacity LN/M/1/N model for different values of traffic intensity.