| Mobile DTNs (Delay Tolerant Networks) represent the combination of the concept of DTN with traditional mobile ad hoc networks, enabling mobile nodes to communicate even in unideal environments where connectivity is only intermittent. Because of the great potential, Mobile DTNs have become a hot research area in recent years. The core of mobile DTNs is to borrow "store-carry-forward" approaches from DTNs, letting data move physically with the nodes carrying them and thus propagate throughout the network in a "mobility-assisted" manner, by exploiting contact opportunities between nodes.In mobile DTNs, communication opportunities are rare and hard to predict, so designing efficient routing protocols is the key challenge. After studying previous work, it is easy to see that desirable DTN routing protocols should be able to replicate data wisely and avoid congestions, essentially making a tradeoff between resource usage and network performance. In the meantime, although mobile DTNs are designed for environments without infrastructure, using infrastructure to enhance DTNs, or using the concept of DTN to extend the reach of infrastructure, may yield great performance improvements.This paper will discuss key techniques in Mobile DTNs, regarding how to route data efficiently and meanwhile avoid network congestions, as well as how to disseminate data in social networks; in addition, this paper will also provide solutions on how to integrate the concept of DTN with communication infrastructure to improve network performance.First, this paper proposes a new congestion aware DTN routing protocol, CADR, which attempts to efficiently use resources and intelligently react to congestions. CADR calculates relay utility based on history of contacts among nodes, and carefully consider multi-hop paths, in order to provide direction for replication, and at the same time, CADR also determines each node’s responsibility for data delivery, based on which resource allocation is optimized. Congestion control is another important issue. The mature end-to-end congestion control mechanism, which is working well in the Internet, cannot be directly applied, because of inherent long delays hindering the propagation of congestion signals to the source. This calls for congestion control mechanisms in which the decisions can be made autonomously with local information only. Thus, CADR tackles this problem via storage management, and tries to avoid unnecessary drops. Our evaluation proves that CADR is better than many classic DTN routing protocols. Its delivery ratio is more than20percent higher, while its overhead is still the lowest; even when network load rises, its performance remains relatively good.Second, this paper proposes a new data dissemination protocol, attempting to achieve high efficiency in mobile DTNs. Data dissemination is useful in many applications, including event notification, network status updates and content publishing. Since mobile devices are usually carried by humans and move with users who carry them, their mobility patterns are highly influenced by users’ social behaviors, which implies analyzing social relationships is a vital means to improve network efficacy. Following this idea, the protocol identifies suitable relays through considering both the social contact patterns and user preferences, in order to convey data to the users who are interested and, at the same time, prevent excessive replications. Combined with storage management, this protocol can achieve good cost effectiveness, which is verified by our trace-driven simulations. In two real scenarios, the protocol achieves the highest network utility and cost effectiveness, and thus is much better than other protocols tested; meanwhile, this protocol can make a balance between network utility and cost effectiveness by tuning parameters.The integration of communication infrastructure and the concept of DTN is another direction for research. In recent years, Wi-Fi hotspots have been widely deployed all over the world for wireless access. Although they provide only limited coverage, the property that they are interconnected with high speed facilitates data distribution in large areas. Inspired by this fact, this paper proposes a new DTN routing protocol, IEDR, which is an expansion of CADR and attempts to use APs (Access Points) for data propagation. The schemes for uploading and downloading data are carefully designed. Our evaluation proves that IEDR achieves what is expected even with a small number of APs. The delivery ratio is increased by over3percent when the node density is high, and the average transfer delay is decreased by about10percent, while the overhead is also reduced by nearly6percent.From another angle, the asynchronous routing scheme in DTNs can also be utilized to extend the reach of infrastructure. Wi-Fi hotspots offer sizable access data rates, exhibit short RTTs, and are usually underutilized. The above characteristics make them desirable for providing Internet access to mobile clients. In spite of its advantages, the main weakness of WLAN technology used by hotspots is the very limited coverage, which is usually no more than hundreds of meters. Users can obtain WLAN service only when they are in range of a hotspot, and thus experience intermittent connectivity when they are moving. To fill in the gaps between APs, this paper presents a new framework, which combines infrastructure composed of APs with the concept of DTN to improve network performance. In particular, when a mobile client moves out of the reach of infrastructure, our framework predicts the chance that this client will encounter other clients, and lets the APs in the proximity identify suitable relay nodes among the mobile clients that happen to visit those APs. If some passing by mobile clients have a chance to encounter the target client soon, the corresponding AP(s) will transfer the data destined to the target client to those mobile clients, in the hope that they will accomplish the delivery in a "store-carry-forward" fashion.The key issue here is relay selection, as simply selecting clients nearby is not efficient. In this framework, the APs filter visiting mobile clients according to heuristic rules, and select suitable clients as relays to carry data to some target clients, allowing a target client to receive data even before it arrives at an AP. In order to predict contacts among mobile clients, two different methods are used. First, APs are usually stationary and can be regarded as landmarks, therefore future movements of mobile clients can be predicted based upon history of encounters between mobile clients and APs; second, the sequences of APs visited by mobile clients are suitable for data mining to extract encounter patterns, and contact possibilities between clients are then able to be inferred from the collective behavior exhibited by all the clients in the past. Our evaluation shows that the framework fairly improves the performance when there are many clients, reducing average transfer delay by20to30percent, while the overhead is still under control. |
PDF Full Text Request