A Machine Learning-Aided Network Contention-Aware Link Lifetime- and Delay-Based Hybrid Routing Framework for Software-Defined Vehicular Networks

Abstract

The functionality of Vehicular Ad Hoc Networks (VANETs) is improved by the Software-Defined Vehicular Network (SDVN) paradigm. Routing is challenging in vehicular networks due to the dynamic network topology resulting from the high mobility of nodes. Existing approaches for routing in SDVN do not exploit both link lifetimes and link delays in finding routes, nor do they exploit the heterogeneity that exists in links in the vehicular network. Furthermore, most of the existing approaches compute parameters at the controller entirely using heuristic approaches, which are computationally inefficient and can increase the latency of SDVN as the network size grows. In this paper, we propose a novel hybrid algorithm for routing in SDVNs with two modes: the highest stable least delay mode and the highest stable shortest path mode, in which the mode is selected by estimating the network contention. We distinctly identify two communication channels in the vehicular network as wired and wireless, where network link entropy is formulated accordingly and is used in combination with pending transmissions to estimate collision probability and average network contention. We use the prospect of machine learning to predict the wireless link lifetimes and one-hop channel delays, which yield very low Root Mean Square Errors (RMSEs), depicting their very high accuracy, and the wireless link lifetime prediction using deep learning yields a much lower average computational time compared to an optimization-based approach. The proposed novel algorithm selects only stable links by comparing them with a link lifetime threshold whose optimum value is decided experimentally. We propose this routing framework to be compatible with the OpenFlow protocol, where we modify the flow table architecture to incorporate a route valid time and send a packet_in message to the controller when the route’s lifetime expires, requesting new flow rules. We further propose a flow table update algorithm to map computed routes to flow table entries, where we propose to incorporate an adaptive approach for route finding and flow rule updating upon reception of a packet_in message in order to minimize the computational burden at the controller and minimize communication overhead associated with control plane communication. This research contributes a novel hybrid routing framework for the existing SDVN paradigm, scrutinizing machine learning to predict the lifetime and delay of heterogeneity links, which can be readily integrated with the OpenFlow protocol for better routing applications, improving the performance of the SDVN. We performed realistic vehicular network simulations using the network simulator 3 by obtaining vehicular mobility traces using the Simulation of Urban Mobility (SUMO) tool, where we collected data sets for training the machine learning models using the simulated environment in order to test models in terms of RMSE and computational complexity. The proposed routing framework was comparatively assessed against existing routing techniques by evaluating the communication cost, latency, channel utilization, and packet delivery ratio. According to the results, the proposed routing framework results in the lowest communication cost, the highest packet delivery ratio, the least latency, and moderate channel utilization, on average, compared to routing in VANET using Ad Hoc On-demand Distance Vector (AODV) and routing in SDVN using Dijkstra; thus, the proposed routing framework improves routing in SDVN. Furthermore, results show that the proposed routing framework is enhanced with increasing routing frequency and network size, as well as at low vehicular speeds.

A Comprehensive Survey on Knowledge-Defined Networking

“…Due to facing of frequent communication interruptions and poor stability inherent in 3D Flying Ad Hoc Networks (FANETs), a DNN-based routing consisting of a 3D two-space division and DNN-based forwarding that yields better performance in terms of packet delivery rate compared to conventional routing protocols is presented in research [111]. Wijesekara et al generate knowledge regarding link lifetimes and one-hop channel delays related to wired and wireless communication channels in a heterogeneous knowledge-defined vehicular network using DNNs, in order to utilize the generated knowledge in a hybrid stable distance and stable delay based OpenFlow compatible adaptive routing algorithm which yields better routing performance in terms of packet delivery ratio, latency, and communication cost [112]. A DNN-based routing framework called "NeuRoute", which predicts a traffic matrix in real-time using a DNN and generates forwarding rules to optimize network throughput for KDN, is presented in [113].…”

“…Deep neural network To replace optimization models for routing [110] Achieve quasi-optimal performance Deep neural network 3D two space division, forwarding for FANETs [111] Better performance in packet delivery rate, energy-saving Deep neural network Hybrid stable delay and distance based routing [112,200] High packet delivery ratio, low latency and communication cost…”

A Review on Deploying Blockchain Technology for Network Mobility Management

“…-In opportunistic communication grounded offloading, alternative communication technologies such as using device to device communication, using unlicensed spectrum, etc. [58]. In Device-to-Device (D2D) communication, gadgets directly communicate among themselves in proximity using technologies such as Bluetooth or peer to peer Wi-Fi to offload traffic from the mobile network.…”

Mitigating Timing Side-Channel Attacks in Software-Defined Networks: Detection and Response

“…Machine learning techniques have been demonstrated to produce accurate results with low false positive rates. Patikiri et al [33] recently proposed a machine learning algorithm to route applications and improve the performance of software-defined vehicular networks. Tayhan et al [34] presented a thorough analysis of the integration of machine learning and deep learning techniques for intrusion detection in SDN.…”