Abstract

Document Sections

I. Introduction

II. Network Scenarios

III. Simulation Results and Discussion

IV. Conclusion

Authors

Figures

References

Keywords

Metrics

Abstract:

Network simulation is a tool used to analyze and predict the performance of Industrial Internet of Things deployments while dealing with the complexity of real testbeds. Large network deployments with complex protocols such as transmission control protocol are subject to chaos-theory behavior, i.e., small changes in the implementation of the protocol stack or simulator behavior may result in large differences in the performance results. In this article, we present the results of simulating two different scenarios using three simulators. The first scenario focuses on the Incast phenomenon in a local area network where sensor data are collected. The second scenario focuses on a congested link traversed by the collected measurements. The performance metrics obtained from the simulators are compared among them and with ground-truth obtained from real network experiments. The results demonstrate how subtle implementation differences in network simulators impact performance results, and how network engineers must consider these differences.

Introduction

The Industrial Internet of Things (IIoT) is adding large numbers of devices to communication networks. In this era, new devices are connected and old ones are removed, equipment from different vendors is added, old applications stop being used and new ones are developed, controller servers are moved from one switch to another or even to a remote site, new links with increased capacity are added, and new routing protocols are deployed, perhaps without removing old ones, all of this at a really fast pace. Any of these changes can yield consequences in network performance that are very difficult to predict. In complex systems, such as modern networks, because of nonlinearities and synchronizations, small changes can yield large-scale consequences. This phenomenon has been referred to as the amplification principle [1].

At present, it is unfeasible to predict network behavior and performance using analytical tools. This is because a large number of elements must be considered in the model, including nondeterministic behaviors (e.g., user actions). Researchers in academia have attempted approximations based on the analysis of complex systems [2] or chaotic systems [3]; however, the results are not applicable for problem solving in real networks, and engineering approximation continues to be employed in industry [4].

An alternative solution to analytical prediction is simulation modeling. New network or system configurations are tested before actual deployment using simulation software. A network administrator can predict the effects of moving localized controller servers to a cloud service by simulating the present environment and then implementing larger delay and loss parameters and network bottlenecks in the access links to simulate the future environment. New network topologies can be tested, and the results of link utilization can be evaluated. Depending on the complexity of the scenario or the expected changes, simple or complex models can be implemented in the simulation.