Balancing latency and cost in software-defined vehicular networks using genetic algorithm
Introduction
Vehicular ad hoc networks (VANETs) improve the driving safety of vehicles through communications between on-board units (OBUs) and roadside units (RSUs) (Rodríguez et al., 2017), (Nguyen-Minh et al., 2016). Software-defined vehicular networks (SDVNs) incorporate the VANET with software-defined networks (SDNs). The SDVN assists the intelligent transportation system (ITS) in overcoming the problems of enormous number of vehicles, big data flow, heterogeneous vehicles, and frequent topology change owing to rapid vehicle movement, in order to increase driving safety and comfort. In addition, the advances in 5G networks (Aujla et al., 2017), (Araniti et al., 2013), (Sharma et al., 2017a) and edge computing (Deng et al., 2017a), (Sharma et al., 2017b) will further improve the network speed and load to effectively enhance VANETs.
Consider an SDVN deployed along a road in Fig. 1, in which the SDN controller is a logic-centralized control center, and has features of programmability and flexibility. The networking functions of RSUs and BSs in the VANET are separated out to be virtualized and to be unified to be managed. RSUs and BSs remain the network accessing function and the flow table to handle tasks of transmission. Those abstract networking functions are centralized to be managed by the SDN controller, which has a global knowledge view, and can remotely control the flow of any packet in the VANET. Such an emerging technology makes up the factors lacked by the conventional VANET to constitute an SDVN (Bizanis and Kuipers, 2016), (Zheng et al., 2016).
In the SDVN, each vehicle communicates with other vehicles and RSUs through the technology of dedicated short-range communication (DSRC). While WiFi is mainly adopted for wireless local area networks (WLANs), which are supported by IEEE 802.11ax (Deng et al., 2017b), (Hang et al., 2017), (Deng et al., 2014), DSRC is intended for wireless communications with high security and high speed between vehicles and the infrastructure, which are used in the physical layer and the MAC layer in IEEE 802.11p (WAVE) and IEEE1609.1/.2/.3/.4 (Lu et al., 2014). Federal Communications Commission allocated 75 MHz of spectrum in the 5.9 GHz band to be used for DSRC. Vehicles outside the DSRC communication range can adopt cellular networks to connect to a base station (BS); and RSUs, BSs, and the SDN controller communicate with each other through wired fiber networks.
The concept of SDN was proposed by Open Networking Foundation (ONF) (Nunes et al., 2014). One of the main features in the SDN is to introduce a control plane and a data plane. The control plane is a logic-centralized control center in charge of making logical decisions of protocols; and the data plane is in charge of accepting the commands issued by the control plane to transmitting packets (Rathee et al., 2017). The technologies on vehicular networks are roughly categorized into the following three modes (Ku et al., 2014): central control mode, distributed control mode, and hybrid control mode. The hybrid control mode is faster than the central control mode, and can handle service requests in a more complete way than the distributed control mode. Therefore, the hybrid control mode is more suitable for SDVNs. Such an approach is more flexible than conventional physical network facilities (Deng and Chang, 1999), (Deng et al., 2008). It only needs to send commends to the control plane to flexibly allocate network resources, and does not need a huge cost to modify hardware facilities.
The control plane has two types of communications: cellular and DSRC networks. Cellular networks can transmit packets efficiently, but may consume a lot of energy. Relatively, DSRC networks cost less, but have higher latency than cellular networks. If only one type of communications is adopted, the SDVN will have more burdens. Therefore, it is a key to find the optimal strategy to balance usage of the two types of communications (Li et al., 2016).
As for the research on reducing latency and cost, Salahuddin et al. (2015) proposed a framework of RSU clouds, including common RSUs and an RSU micro-datacenter. They further proposed a Markov decision process to find the transmission path with the minimal energy consumption cost. Consider the RSU micro-datacenter's computing ability. A good solution is to introduce fog computing to the SDVN. With flexibility and programmability, the SDN controller does not have a full control authority on all packet flows, but it just provides rules to RSUs assisted with fog computing to remarkably save the time of transmitting packets to the SDN controller (Truong et al., 2015). Fog computing can be regarded as a small-scale cloud system at any time around, which may not achieve the same performance with a true cloud system, but can substantially reduce the latency (Hussain et al., 2012), (Lee et al., 2014). Different from distributed frameworks above, He et al. (2016a) proposed a modified constrained optimization particle swarm optimization approach to balance the load of different RSUs, so that RSUs are not overloaded to generate high latency so as to promote quality of service (QoS). Note that QoS has been one of the most crucial measures in evaluating performance of both cellular networks and WLANs (Deng and Yen, 2005), (Deng et al., 2016).
This work aims to balance usage of DSRC and cellular links to concurrently reduce both latency and cost of the whole SDVN. In practice, because the DSRC is regarded as a kind of network-as-a-service (NaaS) (Car 2Car, 2007), the cost of renting DSRC links is paid by the network provider. Differently, the cost of renting cellular links is paid by each vehicle, so that the driver may try not to frequently use cellular links. Therefore, to encourage drivers to more frequently use cellular links to reduce the latency of transmitting packets, this work considers a strategy of rebating the bandwidth of cellular links, so that more bandwidth of DSRC links is available to increase drivers' satisfaction. Therefore, this work first establishes a mathematical model for the problem of optimizing the rebating strategy, and then proposes a genetic algorithm (GA) to solve this problem. The GA is a metaheuristic algorithm for solving combinatorial optimization problems, and has been applied in solving problems in various networks (e.g., (Kaliappan et al., 2016)). Through repetitive improvement of selection, crossover, and mutation, the GA can search for the optimal solution or nearly optimal solution within a limited time period.
This work enjoys the following contributions:
- •
To find the optimal strategy of balancing latency and cost, this work constructs a mathematical model for rebating bandwidth of cellular links used by heterogeneous vehicles.
- •
Different from the previous works that optimized the latency and the cost separately in two stages, this work proposes an improved GA to optimize the two measures in one stage.
The rest of this work is organized as follows. Section 2 gives the preliminary knowledge of this work. Section 3 creates a mathematical model for the problem concerned in this work. Section 4 introduces the proposed GA for solving this problem. Section 5 shows the simulation results, and Section 6 concludes this work.
Section snippets
Preliminaries
This section first reviews the previous works on SDN applied to the VANET, and then introduces optimization problems of the control plane in SDVNs.
Mathematical model
Assume that each vehicle can access the IP network to transmit packets through either DSRC links or cellular links, in which the cost of using DSRC links is paid by the network provider; whereas the cost of using cellular links is paid by vehicles. To encourage vehicles to use cellular links to reduce latency, the SDN controller rebates a different ratio ηi of the bandwidth of cellular links to each vehicle vi. From the aspect of the network provider, if the rebated bandwidth ηi of vehicle vi
Proposed IGA
The proposed IGA is based on the GA (Holland, 1992), which simulates the evolution process of a population of chromosomes to search for optimal solutions. Because the GA has the feature of rapid convergence so that it tends to fall into local optimum, the proposed IGA bases on the dynamic adjustment scheme in (Mahdavi et al., 2007) to improve the mutation operator. The flowchart of the proposed IGA is given in Fig. 2.
The main steps of the proposed IGA is explained as follows:
Step 1. Initial a
Implementation and experimental results
This section first introduces the experimental environment and the assumptions made in the experiment, and then demonstrates the experimental results.
Conclusion
This work has investigated an optimal rebating strategy to balance the latency and the cost in VDVNs. Different from the previous work that optimized them separately, this work has proposed an IGA to optimize them concurrently. The IGA incorporates the classical GA with a dynamic adjustment scheme of mutation rates to ensure solution diversity and avoid premature convergence. In addition, because a large mutation rate to be adjusted damages the population, the IGA keeps the best chromosome
Acknowledgements
The authors thank the anonymous referees for comments that improved the content as well as the presentation of this paper. This work has been supported in part by Ministry of Science and Technology, Taiwan, under Grant MOST 106-2221-E-009-101-MY3.
References (41)
- et al.
Enhancing energy efficiency and load balancing in mobile ad hoc network using dynamic genetic algorithms
J. Netw. Comput. Appl.
(2016) - et al.
An improved harmony search algorithm for solving optimization problems
Appl. Math. Comput.
(2007) - et al.
Reliable broadcasting using polling scheme based receiver for safety applications in vehicular networks
Veh. Commun.
(2016) - et al.
Energy efficient device discovery for reliable communication in 5G-based IoT and BSNs using unmanned aerial vehicles
J. Netw. Comput. Appl.
(2017) - et al.
LTE for vehicular networking: a survey
IEEE Commun. Mag.
(2013) - et al.
Data offloading in 5G-enabled software-defined vehicular networks: a Stackelberg-game-based approach
IEEE Commun. Mag.
(2017) - et al.
SDN and Virtualization solutions for the Internet of things: a survey
IEEE J. Mag.
(2016) - et al.
QoE-based flow management in software defined vehicular networks
- et al.
SDN enabled content distribution in vehicular networks
Car 2 car communication consortium manifesto: overview of the C2C-CC system
Service-oriented dynamic connection management for software-defined internet of vehicles
IEEE Trans. Intell. Transport. Syst.
A priority scheme for IEEE 802.11 DCF access method
IEICE Trans. Commun.
Quality-of-service provision system for multimedia transmission in IEEE 802.11 wireless LANs
IEEE J. Sel. Area. Commun.
Contention window optimization for IEEE 802.11 DCF access control
IEEE Trans. Wireless Commun.
IEEE 802.11ax: next generation wireless local area networks
On quality-of-service provisioning in IEEE 802.11ax WLANs
IEEE Access.
Latency control in software-defined mobile-edge vehicular networking
IEEE Commun. Mag.
IEEE 802.11ax: highly efficient WLANs for intelligent information infrastructure
IEEE Commun. Mag.
CR-SDVN: a cognitive routing protocol for software-defined vehicular networks
IEEE Sensor. J.
Performance analysis of IEEE 802.11ax UL OFDMA-based random access mechanism
Cited by (22)
A fault-tolerant adaptive genetic algorithm for service scheduling in internet of vehicles
2023, Applied Soft ComputingExploring software defined networks for seamless handovers in vehicular networks
2021, Vehicular CommunicationsCitation Excerpt :The core logic is to build a cluster of nodes that follow similar social matches, which can possibly have the same future routes. An improved genetic algorithm has been also proposed to optimize the dynamic network changes in VANETs [19]. It includes the solution for adjusting the dynamic changes often occurring in the network, and also gives the guarantee about the solution diversity.
Security and design requirements for software-defined VANETs
2020, Computer NetworksCitation Excerpt :Specifically, SDN’s programming flexibility feature provides great support for implementing fog computing services at SDN-enabled edge devices. For instance, authors in [103] propose a scheme for balancing service latency and cost by using genetic algorithms in SDVNs, and authors in [35] proposes a centralized routing scheme with mobility prediction for SDVNs to minimize the overall vehicular service delay. It also uses an artificial intelligence-powered SDN controller.
Network selection and data dissemination in heterogeneous software-defined vehicular network
2019, Computer NetworksCitation Excerpt :An optimal rebating strategy is used to balance the cost of 5G cellular network and latency of the vehicular network. In the same context, Lin et al. [32] proposed a rebating strategy to balance latency requirement and the cost using an Improved Genetic Algorithm (IGA) in SDVN. The IGA ensures the best solution and avoids premature convergence by modifying the simple GA using dynamic self-adjusting mutation.
Multipath TCP for V2I communication in SDN controlled small cell deployment of smart city
2019, Vehicular CommunicationsCitation Excerpt :The proposed scheme considers both the quality of the network and the estimated residence time of a vehicle in the RSU to make a decision. In the literature [24–27], the researchers have also tried to explore how emerging technologies such as fog computing, network function virtualization (NFV), mobile edge computing (MEC), named data network [28] and artificial intelligence can play a role in the SDVN to facilitate quality of service and quality of experience requirements. We found two good studies [29,30] that attempted to address the issues arose due to vehicular mobility in SDVN.