7.3 Related Works on Mobile Content Distribution in VANETs
7.3.1 Cooperative Downloading of General Contents in VANETs
In (Nandan et al. 2005), Nandan et al. first studied cooperative downloading in VANETs. They proposed SPAWN, a pull-based, peer-to-peer content downloading protocol for VANETs that extends BitTorrent. Later, they proposed “AdTorrent” (Nandan et al. 2006), which is a semi-push-based peer-to-peer protocol for vehicles to download advertisements they are interested in. In both SPAWN and AdTorrent, the peer and content selection mechanisms have a high overhead and are not scalable, especially when most of the vehicles are interested in downloading popular contents. Also, they suffer from the “coupon collector problem,” which enlarges downloading delay. Moreover, they use Transport Control Protocol (TCP) for content delivery, which performs poorly over multihop lossy wireless links in highly mobile VANETs.
7.3.1.1 Network Coding for Content Downloading
To avoid such problems, many researchers resort to network coding (Ahlswede et al. 2000; Ho et al. 2006). Lee et al. proposed CodeTorrent (Lee et al. 2006), a pull-based content distribution scheme using network coding, where vehicles need to explicitly initiate requests to download a piece of content. CodeTorrent restricts the peer selection and content delivery to the one-hop neighborhood of a vehicle, thus eliminating the need of multi-hop routing. Also, the use of network coding mitigates the peer and content selection problems.
Lee et al. (2008) further studied the practical effects of content distribution in VANETs using network coding based on a variation of CodeTorrent. It is shown that the resource constraints such as disk access, computation and buffer have significant impacts on the performance. They discussed approaches to reduce the communication and computation overhead of network coding while maintaining the gain of it.
The above schemes are all pull based in essence. They could suffer from large downloading delay, because nodes passively respond to their neighbors' requests and the bandwidth is wasted (being idle much of the time). For example, in CodeTorrent it takes 200 seconds to download a 1 MB file in an urban scenario (Lee et al. 2006). If a node wants to receive new information continuously, it must send out requests frequently. The transmissions from multiple responders tend to collide with each other, leading to low efficiency in turn. Park et al. proposed a push-based content delivery scheme for emergency related video streaming using network coding (Park et al. 2010; 2006). However their “push” protocol design essentially reduces to controlled flooding, which tends to be inefficient.
In fact, with packet-level network coding (PLNC), it is difficult to achieve high downloading performance especially under lossy wireless links in VANETs, whether or not a push-based protocol design is adopted. The wireless medium in VANET has been shown to be lossy by empirical analysis (Ramachandran et al. 2007; Taliwal et al. 2004; Torrent-Moreno et al. 2006). In practice, network coding for a large file is usually done within each block of the file, namely a generation (Ahmed and Kanhere 2006; Lee et al. 2006; 2008). In order to maintain reasonable coding/decoding complexity while reducing the protocol overhead, the basic coding unit (coded piece) should be larger than a usual packet. During the transmission of such a coded piece, any error to the coding vector or message body will render the whole piece useless, leading to degraded downloading performance. This is part of the reason that we resort to SLNC (Katti et al. 2008), which has much better resiliency to transmission errors due to symbol-level diversity.
7.3.1.2 Transmission Coordination in Content Downloading
Transmission coordination is an important issue for MCD in VANETs. Bad coordination could result in severe packet collisions that affect downloading performance. However, this issue has not been well addressed in previous works. In Park et al. (2006), a simple time-out mechanism is used for each vehicle to decide when to transmit a coded packet. However, this mechanism does not take into account vehicles' content reception status, which leads to a non-negligible chance of duplicate information. Packet collisions are severe when the network is dense.
Zhang et al. (2009a) studied this problem from a link-layer perspective, and proposed VC-MAC, a cooperative medium access control (MAC) protocol for gateway downloading scenarios in vehicular networks. In order to avoid possible interference among multiple transmissions and to maximize the “broadcast throughput,” a heuristic relay selection algorithm with a backoff mechanism is proposed. However, the “broadcast throughput” is purely based on link quality, which is not content aware. The relay chosen by VC-MAC may have nothing innovative to transmit to its neighbors.
In CodeOn in this chapter, we explicitly consider the content usefulness of nodes for higher rate content downloading. A dynamic set of relay nodes, which are selected based on their content availability and usefulness, actively broadcast (push) useful contents to neighboring nodes and make medium access decisions based on both their content usefulness and local channel status.
7.3.1.3 Multichannel Compatibility
There has been little consideration of the compatibility of content downloading with other channels. In Mak et al. (2009), the authors propose mechanisms to adjust the time share of the service channel to enhance the performance of content downloading while guaranteeing the quality of service (QoS) of safety messages. In this chapter, we consider the coexistence of a service channel with the control channel, with the difference that we design a better PCD protocol given a fixed time share of service channel. We also use the control channel for better content downloading.
7.3.1.4 Other Related Works
Zhao et al. (2007) proposed data pouring, a push-based data dissemination protocol for VANETs. They focus on broadcasting small data items to all vehicles inside an area, while we aim at disseminating large popular files. Zhao et al. (2008) also studied the problem of drive-thru access to roadside APs, and proposed a vehicle-to-vehicle relay strategy to extend the coverage of APs. Yang et al. (2009), proposed a push-based, reliable broadcast protocol for wireless mesh networks using network coding.
In addition, Fiore and Barcelo-Ordinas (2009) focused on cooperative downloading in urban VANETs. The Roadcast (Zhang et al. 2009b) is a popularity-aware content-sharing protocol in VANETs. These protocols are mainly suitable for applications where each vehicle may be interested in downloading different files, while here we consider the broadcast of popular content.
7.3.2 Streaming of Multimedia Content in VANETs
Currently, streaming services such as PPLive and PPStream are widely used on the Internet. In particular, network coding has been shown to be an effective technique that can improve the user experience of video streaming service for large-scale systems. For example, Wang and Li (2007) proposed R2, a random push-based P2P scheme using network coding. Also, Zimu Liu (2010) deployed a NC-based on-demand streaming scheme in a large-scaled commercial system which showed the benefits of NC for multimedia streaming in a real P2P network. In wireless mesh networks, Seferoglu and Markopoulou (2009) proposed a video-aware opportunistic network coding scheme across different flows. However, all these schemes are for traditional wired or wireless networks and are not suitable for VANETs, due to VANETs' unique characteristics described previously.
For VANETs, Bonuccelli et al. (2007), Bucciol et al. (2005), Guo et al. (2005), Park et al. (2006), Qadri et al. (2009) and Soldo et al. (2008) proposed several schemes for supporting various kinds of streaming services, which can be divided into two categories:
1. Schemes focusing on the application layer. Bonuccelli et al. 2007 proposed a real-time video transmission scheme in vehicular networks. This scheme only considers unicast sessions and relies heavily on fast and reliable feedback from the receiver side. Bucciol et al. (2005) carried out a series of experiments using two vehicles under different scenarios, which proved the feasibility of video streaming between moving vehicles. Qadri et al. (2009) showed that by adopting error resilience coding, state-of-the-art routing protocols can support multicast video streaming in city VANETs when the network is not dense. These works mainly showed the feasibility of video streaming in VANETs and have not considered more practical issues such as dealing with dynamically changing network density and minimizing bandwidth cost, which are taken into considered in this chapter.
2. Schemes focusing on network and MAC layer. Park et al. (2010) proposed NCDD for emergency related video streaming in VANETs using NC. In this scheme, the transmission of each vehicle is triggered by a timer set upon the reception of every new packet. Since neighbors' current reception status is not considered, the broadcast packets are not always useful for nodes' neighbors, which decreases the bandwidth efficiency. Due to lack of coordination between concurrent transmitting vehicles, the scheme also tends to suffer from severe collisions, especially under dense vehicular traffic.
Soldo et al. (2008) introduced SMUG, a TDMA-based scheme to support streaming media dissemination in city VANETs. A tree structure is established for broadcasting streaming video content. However, it is hard to maintain a stable and up-to-date communication structure for dynamic VANETs, thus stable streaming rate is difficult to achieve. Guo et al. (2005) proposed V3, a live video architecture for VANET, where directed broadcast is adopted for remote video request scenarios, which are different from the application in this chapter.