Vehicular Delay Tolerant Network (VDTN): Routing Perspectives Syed Hassan Ahmed1 , Hyunwoo Kang2 and Dongkyun Kim3 1,3
2
School of Computer Science & Engineering, Kyungpook National University, Daegu, Republic of Korea. Electronics and Telecommunication Research Institute (ETRI), Republic of Korea. Email: {1 hassan,3 dongkyun}@knu.ac.kr,2
[email protected]
Abstract—Recently, the Delay Tolerant Networks (DTN) have been utilized in various operational communication paradigms. This includes the communication scenarios that are subject to disruption and disconnection as well as the scenarios with high delay and frequent partitioning, i.e., Vehicular Ad hoc Networks (VANETs). Due to several characteristics match, a new research paradigm named as Vehicular Delay Tolerant Network (VDTN) is introduced. Through relays and store-carry-forward mechanisms, messages in VDTNs can be delivered to the destination without an end-to-end connection for delay-tolerant applications. However, the choice of routing algorithms in VDTNs is still under study. Numerous routing protocols have been proposed to meet requirements of many applications. In this paper, we therefore provide some detailed study of recently proposed routing schemes for VDTNs. We also perform comparative analysis on the basis of unique criterion such as forwarding metrics with their implementations. In addition, open challenges and future directions are provided to make room of interest for the research community1 . Keywords—VDTNs, V2V, V2I, Location Information, Routing.
I.
I NTRODUCTION
Vehicular Ad-Hoc Network (VANET) is an on-demand network of vehicles that holds the features such as selforganization, auto-configuration, and self-healing. The VANET topology can vary from really dense (e.g., rush hour, traffic jams, etc.) to very sparse (e.g., rural area, late night, etc.). In case of dense network topologies, it is easy to provide an end-to-end multi-hop communication between source and destination vehicles due to the presence of vehicles in the path. Similar communications between sender and receiver vehicles cannot be achieved in sparse vehicular network scenarios because of the intermittent nature of intermediate links. The communications between vehicles in sparse vehicular network case are achieved by using the store-carry-and-forward (SCAF) communication mechanism that is the foundation of Delay Tolerant Network (DTNs)[1]. This type of vehicular network paradigm is called Vehicular Delay Tolerant Network (VDTN).
that are vehicles in this scenario. Generally, the bundle protocol does not provide details of routes for data packets between the nodes. It deals only with the forwarding phase. Since enabling end-to-end connectivity in vehicular networks is a significant issue and needs to be addressed by appropriate routing approaches, a number of studies have been carried out for applicable routing protocols for VDTN based on different schemes, such as model-based schemes, epidemic schemes and estimation schemes [2]. Several routing protocols have been proposed for VDTN with an attempt to achieve high delivery ratio, minimum delay at minimum bandwidth, storage, congestion, and exchange of control information. A very simple protocol is Direct Delivery, in which the node originating a message carries it until it meets its final destination. In First Contact routing, the nodes forward messages to the first node they encounter, which results in a ”random walk” search for the destination node. Epidemic routing replicates messages to all encountered peers that still do not have them [3]. If a message storage space is unlimited and contacts between nodes are long enough, the epidemic routing minimizes the delivery delay and maximizes the delivery ratio. However, since those resources are usually limited, the epidemic routing wastes storage and bandwidth compared to other protocols. For instance, Surround routing tries to minimize the storage consumption and overhead by also sending messages to all the nodes, but only the nodes that surround the final recipient will keep the copies longer than others. Spray-and-Wait [4] generates n copies of a message. In a normal mode, a node gives one copy to each contact; in a binary mode, half of the copies are forwarded to a contact. Once only a single copy is left, it is forwarded only to the final recipient. Spray-and-Wait is another example of protocols that limit message replication as compared with Epidemic routing. The PRoPHET (Probabilistic Routing Protocol using History of Encounters and Transitivity) [5] protocol transfers the message to a neighbor if it estimates that the neighbor has a higher ”likelihood” of being able to deliver the message to the final destination based on the past node encounter history.
In DTNs, a message-oriented overlay layer called ”Bundle Layer”, above the transport (or other) layers, is used to transform application data units into one or more protocol data units called ”bundles”. The DTN nodes use the Bundle Protocol to communicate bundles between source and destination nodes
Hence, we conclude that vehicular DTNs have been investigated for different applications with a large number of proposed routing algorithms.
1 The extended version of this manuscript is submitted to International Journal of Distributed Sensor Networks.
From the literature, we can easily find out some quality survey papers in various areas of VANETs [6]. However,
A. Motivation
the focus of those surveys is mostly built around routing issues in VANETs without taking DTN characteristics into account. Later, some authors in [7] took initiative to provide the performance of VDTNs routing protocols. However, any comparative analysis has not been performed. Hence, the current literature still lacks in thorough studies providing more insight on the routing issues in vehicular DTNs. In this paper, we therefore provide a comprehensive review of Vehicular Delay Tolerant Networks (VDTNs) routing protocols. Furthermore, we also perform a comparative analysis of selected protocols while defining some metrics such as forwarding metrics, infrastructure-assisted, location-information, topology assumptions, implementation and main objectives. Moreover, we summarize the future research directions in this demanding paradigm.
Fig. 1.
VDTN in PBRS
Fig. 2.
Communication in ASCF
The rest of this article is organized as follows. In Section II, we present the detail of selected DTN routing schemes proposed for VANETs. Section III provides the comparative analysis of vehicular DTN protocols. Open issues and future directions are given in Section IV. Finally, Section V concludes this article. II.
VDTN ROUTING P ROTOCOLS
A. PBRS (Probabilistic Bundle Relaying Scheme) The Road-Side Units (RSUs) support communications between vehicles and infrastructures for numerous applications. However, in real environments, RSUs cannot cover all the roadside areas because of the deployment cost. Thus, communications over relaying vehicles are considered one of the solutions to support the uncovered areas by RSUs. Some typical researches utilized store-and-forward techniques for relaying data between RSUs and vehicles. The RSU transmits its data to the incoming vehicles which enter its transmission range. In this case, if an RSU transmits its data to all the vehicles which are passing by it, a lot of replicated packets are generated in the network. Therefore, PBRS [9] proposed a decision-based scheme which makes RSUs determine whether or not to release its data to a vehicle on the basis of certain criterion. Figure 1 shows the vehicular delay-tolerant network which is considered in PBRS. The source RSU S has data to forward to the destination RSU D. However, there is no end-toend path between S and D. The Vehicles passing by S make S become aware of the speed of those vehicles. PBSR calculates the release probability by utilizing the speed of vehicles. When a vehicle Vi enters a communication range of S, the S holds its data until the vehicle moves out of the range or a next vehicle Vi+1 enters the coverage area. If the Vi+1 is faster than Vi and Vi+1 is considered to reach D before the Vi does, S transmits its data to Vi+1 . B. ASCF (Adaptive Carry-Store Forward) ACSF also assumes that RSUs cannot cover all the roadside areas like PBRS. ACSF utilized a store-and-forward technique for relaying data. However, it focused on the outage time of a target vehicle in an uncovered area. In the ACSF scheme, a message forwarding mechanism was proposed for reducing the outage time for vehicles [10]. Figure 2 shows the deployment of vehicles and RSUs considered in ACSF. The authors implemented ACSF for two RSUs partially deployed and leaving
uncovered area between them. Here, the uncovered area means the road segment which is not in the transmission range of any RSU(s). In Figure 2, it is shown that the vehicles move from left to right side of the road. After the entrance of V0 in the covered area of RSU1 , it starts communicating data with RSU1 . Since V0 is moving, after some time it will be entering into the uncovered area. However, the vehicles Vj and Vk can still be used as a relay to receive the remaining data from RSU1 and forward it to V0 . For this purpose, RSU1 selects the node which provides longer connectivity to V0 , thus decreasing the outage time. The outage time can be calculated by the moving speed of each node. Since RSU can be easily aware of its transmission range and the moving speed of nodes moving in it, RSU can calculate when V0 moves out of its communication range. Before node V0 leaves the coverage area, RSU1 selects the relay node with a maximum connectivity time with V0 . ACSF assumes that V0 is required to adjust its speed in an uncovered area for a longer connection with a relaying vehicle selected by RSU1 . C. DARCC (Distance-Aware Routing with Copy Control) The routing decision aims to determine how to replicate or forward message copies to suitable nodes [11]. DARCC applied this concept of DTN routing to vehicular environments. The vehicles in DARCC determine whether or not to transmit data to their encountering vehicles with two principles. If the location of the destination of data is available, the data is forwarded to the vehicle that is closer to the destination. Otherwise, DARCC prefers spreading the data to different direction to increase the probability to meet a destination. Figure 3 shows the concept of DARCC. Each vehicle in DARCC is equipped with a GPS, thus the vehicle can calculate its current motion vector. The motion vector is the speed of a vehicle and its moving direction. If the vehicle A turns left in junction during a certain time t, its motion vector of
Routing for Vehicular Networks (GeOpps) aims to enhance the performance of single-copy routing protocol in VDTNs [13]. It exploits the geo-location of vehicles to forward the geographical bundle opportunistically towards the final destination location. Thus, the vehicle that is heading towards or near the destination location of the bundle becomes the next bundle carrier. The closest point where a vehicle carries the bundle is called nearest point, and used to compute Minimum Estimated Time of Delivery (METD) as follows: Fig. 3.
Concept of DARCC
M ET D = time to nearest point +
remaining distance (1) average speed
A vehicle with the lowest METD is the candidate bundle forwarder/carrier. GeOpps assumes that the bundle carrier always find another vehicle when it arrives at the nearest point. In some cases, it might be practical to handover bundle(s) to the vehicle moving slowly to a destination rather than the vehicle that will just reach the nearest point faster. To achieve this, GeOpps assigns weights according to varying speed of vehicles and their remaining distances to the nearest points. However, it does not provide a method to optimally calculate these weights. F. GeoSpray (Geographical Spray in VDTN)
Fig. 4.
Network model in DAWN
the time t is a calculated like arrow in the Figure 3. Each vehicle periodically broadcasts a beacon message including its location, current motion vector and the list of the messages it has. If the vehicles are moving in different directions, the replication helps to perform the successful delivery, because the other vehicles may reach its destination on the move. Thus, the vehicles A and B replicate their packets to each other, respectively. D. DAWN (Density Adaptive routing With Node awareness) The authors in DAWN [12] proposed a data forwarding strategy based on the density of the network. The network scenario is shown in Figure 5 where vehicles and mobile nodes move around randomly in the area. In case of low density in an area, epidemic routing is implemented. Otherwise, throughput is restricted to avoid congestion. In this scenario, DAWN introduces the Utility Incremental Value (UIV) which assigns priorities to the packets. The packets with higher UIVs should be transmitted with higher priority. The UIV is estimated by each node to maximize the probability of packets to be delivered to the base station before a dead line. E. GeOpps (Geographical Opportunistic Routing) Geographic routing is one of the most promising approaches for efficient routing, which takes location information of the vehicle into account. Geographical Opportunistic
GeoSpray [14] uses the principles of single-copy singlepath GeOpps to perform multi-copy multi-path bundle routing approach. Multi-copy routing schemes are noted for their high delivery ratios, low bundle delivery delays, and high overheads due to duplicated copies. Thus, GeoSpray adopts the replication approach of the Spray and Wait protocol [7] to limit the number of copies. Initially, it uses a multiple copy scheme, which spreads a limited copies of the bundle to exploit diverse paths. Afterwards, it switches to a single-copy forwarding scheme. GeoSpray clears the delivered bundles from vehicles’ storage by propagating the delivery information. As a result, it achieves better delivery ratio than GeOpps at the cost of high replication overhead. However, this overhead is less than the Epidemic protocol and similar to Spray and Wait. III.
C OMPARATIVE A NALYSIS
In this section, the comparative analysis of the previously discussed VDTN routing protocols is presented. We compare and analyze the above mentioned schemes based on the following metrics: A. Forwarding Metric Most of VDTN routing protocols utilize the store-carry and forward mechanism. Hence, these protocols usually do not make any end-to-end path between source and destination vehicles. In Epidemic routing which is one of the most famous store-carry and forward routing, the vehicles replicate all the data they have to the all vehicles they encounter. However, in above mentioned schemes, the vehicles which have data should determine whether or not to forward data to encountering nodes with some criteria. Therefore, we define these criteria as forwarding metrics in VDTNs. The forwarding metric is one of the most significant features for distinguishing routing protocols.
TABLE I.
C OMPARATIVE A NALYSIS OF VDTN ROUTING S CHEMES
Scheme Name
Forwarding Metrics
Infrastructure Assisted
Location Information
Topology Assumption
Implementation
Target
PBRS [9]
Velocity-based probability
Yes
Yes
20 km one way road Vehicle inter arrival time : 5-120 seconds
Java-based simulator
Reduce packet replication
ACSF [10]
Minimum outage time of node
Yes
Yes
Not Available
Numerical analysis
Maximum connectivity
Opportunistic Network Environment (One) simulator
Reduce packet replication
Simulation with real environment data
Optimize channel usage
OMNet++
Optimize delivery ratio, delay and overhead
VDTNsim
Optimized routing with minimum delay
DARCC [11]
Location of destination Moving direction of nodes
Yes
Yes
DAWN [12]
Density of nodes
No
Yes
GeOpps [13]
Density of nodes
Yes
Yes
GeoSpray [14]
Density of nodes and Different Data Size
Yes
Yes
100 vehicles in 3000m x3000m area Each road has 4 lanes Average speed of node 60km/h 5000 taxi in Beijing city 30 days of trace 25x25 Manhattan Grid 260,000 vehicles 15km x 15km area 100 mobile nodes with an average speed of 50 km/h city of Helsinki, time: 6hrs
B. Infrastructure Assisted
E. Implementation
As mentioned in Section I, the infrastructures such as RSUs have been installed to support the vehicle-to-vehicle (V2V) communications for increasing reliability, reducing transmission delay, etc. Therefore, some VDTN routing protocols assumed that the infrastructures can support the V2V communication in a whole roadside area, thus improving routing performances. However, this assumption is impractical since the installation of infrastructures costs so much. In the real world, the infrastructures are installed in limited roadside areas and they can support the V2V communications within their coverage2 . Therefore, some VDTN routing protocols such as [11] are designed to be well-operated without any support of the infrastructures. Moreover, some VDTN routing protocols assumed that the support of infrastructures can be provided in the limited area. Since the routing performance depends on the existence of infrastructure, it is an important metric when analyzing VDTN routing protocols.
The efficiency of routing protocols is measured through simulations, numerical methods, implementations or real-world test-bed environment. Here, we classify the VDTN routing protocols according to the testing environment where they have been implemented or tested.
C. Location Information A vehicle in VDTN uses GPS information to perform bundle forwarding decision from the source to destination vehicles. Location information is used by the source vehicle to forward bundle(s) to the encountered vehicle that is nearer to the destination vehicle. Moreover, in some VDTN routing protocols, not only the GPS information but also map information is used to determine an optimal next forwarder. Therefore, the way in which the location information is used is a promising metric to classify VDTN routing protocols. D. Topology Assumptions Beside the dependency on existing infrastructures, VDTN routing protocols may also assume topology parameters e.g., network model, mobility model, traffic characteristics, etc. In addition, although some VDTN routing protocols can achieve some performance improvement over the particular topologies, they cannot achieve the same over another topologies. 2 The localization of the RSUs is still a part of research but out of scope in this manuscript.
F. Target In common VDTN routing protocols, when the source node meets another node (namely encounter node), it should determine whether or not to transmit its packet to the encounter node. At this point of time, the source node calculates a ”cost” based on forwarding metric which is described in Section 3-A. The source node transmits its packet if the cost of encounter node is low. Hence, the forwarding metric can represent the target of routing protocol, but it is not at all times. For example, when the source node wants to transmit its packet to the destination as soon as possible, the speed of encounter vehicle can be used as the forwarding metric. In addition, even if the source node wants to maintain connectivity with the selected encounter vehicle, the speed of encounter vehicle also can be used as the forwarding metric. Therefore, not only the forwarding metric but also the target of protocols is an important metric to analyze VDTN routing protocols. Table I shows the comparative analysis of the VDTN protocols discussed in this survey. In PBRS, a velocity of node is utilized to calculate the release probability. If several nodes are in the communication range of an RSU, the node with higher speed tends to reach destination faster than slower speed node. For this reason, faster nodes get higher release probability in PBSR. Thus, we call this forwarding metric velocity-based probability in the table. Similarly, in ACSF [9], the maximum hop counts is two between the source and destination. The only RSUs are the only source nodes in this scheme. Due to the limitation of communication range, the connectivity between an RSU and a vehicle cannot be maintained. In order to overcome this problem, a relaying vehicle is selected. When multiple vehicles are available for relaying, the one which can maximize the connectivity is selected. The velocity of relaying vehicle and target vehicle is important factor to keep the connectivity. Unlike PBRS, the fastest node is not important in ACSF, because it is easy to
maintain the connection if the speeds of the two nodes are similar. DARCC [11] and DAWN [12] utilize packet replication mechanism like Epidemic routing. Packet replication is a useful technique to increase delivery ratio in DTNs, but it may result in a waste of network resources. Thus, to control the amount of replicated packets appropriately is a significant issue in these protocols. First, DARCC assumes two situations. If the location of the destination is available, the data is forwarded to the vehicle that is closer to the destination. The data is forwarded to the node which are moving in different direction to spread the data over a wide area with a small number of replicated packets. On the other hand, DAWN focuses on the density of nodes in the cell. If the density increases, the congestion also increases. DAWN reduces packet replication only if the channel is congested. It tries to maximize the local channel capacity if the throughput does not fall due to the congestion. IV.
VDTN: R ESEARCH C HALLENGES
In this section, we describe open issues and challenges for VDTN routing protocols. The need to address the emerging number of services in the vehicles has given rise to an increase in research in vehicular communications. The key challenge is routing due to the dynamic topology changes. Many protocols have been discussed in the previous sections. However, there still exist some challenges and open issues that need to be investigated. •
Most of routing protocols assume either the highway scenario or the urban scenarios. The protocol which is designed for such environments may not show efficient performance in a more complex environment. For example, the vehicles may enter the urban area after passing highway. Therefore, variety environments should be taken into account at the same time.
•
In most studies for VDTNs, the buffer management of vehicles is overlooked. Only the size of buffer is described, but how it can be managed is not described. The buffer management is important in DTN, because a lot of DTN protocols are based on store-carry and forward mechanism. Therefore, reallocating buffer space and maximum use of other resources can also be addressed.
•
Most of the routing protocols utilize the location information of nodes. The location information acquisition is not easy when the destination node is mobile. For the stationary node, every node is aware of the nodes location. Hence, location information can be an efficient metric for routing. Another issue is the implementation of the routing protocols in the real world scenarios. Better performance can be predicted from the protocols applied in the real world scenarios.
•
There always exists a tradeoff between delivery ratio, end-to-end delay, and network resource usage while applying different approaches in the vehicular networks. Thus, a completely different algorithm with existing methods can be expected to minimize the tradeoff through, for example, Artificial Intelligenceaware routing.
V.
C ONCLUSION
This paper provides a detailed survey of recent developments in vehicular DTNs with emphasize on routing. To the best of our knowledge, this is the first work to present the comparative analysis of selected Vehicular DTN (VDTNs) routing protocols with respect to unique metrics such as implementation, infrastructure assisted, etc. In addition, a number of open challenges and future directions have been discussed to motivate further research interest for existing routing constraints in VDTNs. ACKNOWLEDGEMENT This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (2012R1A1A4A01009954). This research was also supported by the MSIP (Ministry of Science, ICT & Future Planning), Korea, under the C-ITRC (Convergence Information Technology Research Center) support program (NIPA-2014H0401-14-1004) supervised by the NIPA (National IT Industry Promotion Agency). R EFERENCES [1] Sagar, S., S. H. Ahmed, Z. A. Khan, U. Qasim, I. I. Awan, M. Y. Durani, S. Ahmed, M. D. Sabir, and N. Javaid. ”Link and Path Duration of Routing Protocols in Mobile Ad-hoc Networks and Vehcular Ad-hoc Networks.” (2014). [2] Paulo R. Pereira, Joel J. P. C. Rodrigues and Joan Triay, ”From Delay-Tolerant Networks to Vehicular Delay-Tolerant Networks,” IEEE Communications Surveys & Tutorials, Vol. 14, No. 4, 4th Quarter 2012. [3] A. Bujari, C. E. Palazzi, D. Maggiorini, C. Quadri and G. P. Rossi, ”A Solution for Mobile DTN in a Real Urban Scenario,” WCNC 2012 Workshop, April 2012. [4] Kumar, S., S. H. Ahmed, U. Qasim, Z. A. Khan, N. Amjad, M. Q. Azeem, A. Ali, M. J. Ashraf, and N. Javaid. ”Analyzing Link and Path Availability of Routing Protocols in Vehcular Ad-hoc Networks.” (2014). [5] Xing, Kai, Weili Wu, Ling Ding, Lidong Wu, and James Willson. ”An efficient routing protocol based on consecutive forwarding prediction in delay tolerant networks.” International Journal of Sensor Networks 15, no. 2 (2014): 73-82. [6] Isento, Joao, Joel Rodrigues, Joao Dias, Maicke Paula, and Alexey Vinel. ”Vehicular Delay-Tolerant Networks? A Novel Solution for Vehicular Communications.” Intelligent Transportation Systems Magazine, IEEE 5, no. 4 (2013): 10-19. [7] Zhu, Ying, Bin Xu, Xinghua Shi, and Yu Wang. ”A survey of socialbased routing in delay tolerant networks: positive and negative social effects.” Communications Surveys & Tutorials, IEEE 15, no. 1 (2013): 387-401. [8] Silva, Aloizio P., Scott Burleigh, Celso M. Hirata, and Katia Obraczka. ”A Survey on Congestion Control for Delay and Disruption Tolerant Networks.” Ad Hoc Networks (2014). [9] Khabbaz, Maurice J., Wissam F. Fawaz, and Chadi M. Assi. ”Probabilistic bundle relaying schemes in two-hop vehicular delay tolerant networks.” Communications Letters, IEEE 15, no. 3 (2011): 281-283. [10] Wu, Di, Gang Zhu, and Dongmei Zhao. ”Adaptive carry-store forward scheme in two-hop vehicular delay tolerant networks.” Communications Letters, IEEE 17, no. 4 (2013): 721-724. [11] Lo, Wei-Zen, Jhih-Siao Gao, and Shou-Chih Lo. ”Distance-aware routing with copy control in vehicle-based DTNs.” In Vehicular Technology Conference (VTC Spring), 2012 IEEE 75th, pp. 1-5. IEEE, 2012. [12] Fu, Qiao, Lin Zhang, Wei Feng, and Yixin Zheng. ”Dawn: A density adaptive routing algorithm for vehicular delay tolerant sensor networks.” In Communication, Control, and Computing (Allerton), 2011 49th Annual Allerton Conference on, pp. 1250-1257. IEEE, 2011.
[13]
Leontiadis, Ilias, and Cecilia Mascolo. ”Geopps: Geographical opportunistic routing for vehicular networks.” In World of Wireless, Mobile and Multimedia Networks, 2007. WoWMoM 2007. IEEE International Symposium on a, pp. 1-6. IEEE, 2007. [14] Soares, Vasco NGJ, Joel JPC Rodrigues, and Farid Farahmand. ”GeoSpray: A geographic routing protocol for vehicular delay-tolerant networks.” Information Fusion 15 (2014): 102-113.
