The overlay network approach allows the creation of several virtual networks over physical ones. Here one hop in a virtual network may correspond to several hops from one or more physical networks (underlays). In turn, a virtual network may even be built over other virtual networks. However the overlay routing does not have direct control over how the underlay forwarding of packets is actually performed. As a result one may not classify solutions such as Buble and SimBet as mechanisms to build overlay networks, since their nodes have routing control over the physical layer. Other routing algorithms found in widely spread Peer-to-Peer (P2P) social networks such as Gnutella, Chord and Tapestry are examples of peer-to-peer protocols that can be used to create overlay networks. In order to improve these overlay networks, and other works applied social mechanisms and consequently we named these proposals as social overlay networks. They deal with security and optimization using social ideas extracted from the underlays.
DSL (Davis Social Links) is an example of a proposal that applied social approaches to deal with security. It has a social overlay mechanism that creates trust relationships of social networks based on the small-world phenomenon in order to control, trace and separate address and identity information. The idea is to allow communication between nodes with a direct link or a social path linking them. Messages in the DSL social network contain a set of keywords describing node properties. The social path is created by the exchange of keywords between nodes connected by direct links. The nodes can accept or drop messages from some social path or modify these along another one. There is consequently a recipient controllability being exerted. For example, nodes A and B have "red" and "yellow" as part of their keyword sets respectively. These strings are cryptographically exchanged prior to the exchange of data. A Node can propagate the keywords in order to increase routability.
Work is a bit similar to social routing while not applied at the physical layer however. SOLAR, seen earlier, imports a user's profile information to forward data. In SOLAR, the profile is defined by a node itself and describes node mobility and its likely locations. Differently from SOLAR, acts at the application layer where the profile manipulates the dissemination and routing of messages. An even greater difference is to do with the fact that the profile in may be updated and created by friends and acquaintances of a given node instead of being the sole responsibility of the node as in SOLAR. When a node creates a link, the user must attribute one or more keywords to describe the new friend. This one, the friend, needs then to agree with such description to allow for the effective link creation. As described by an application in, it is important to find a best person to answer some questions about some given keywords. This idea for creating profiles by friends could be implemented at the network level in terms of performance and made available to view by other nodes. Consequently, a profile that describes the options and possibilities for other nodes could be an interesting mechanism to help the node decide if another encountered node is a good relay to forward its messages.
PeopleNet is another social mechanism that propagates information using overlay networks. It has only three types of messages: request propagation, request and response. The request and response messages are always forwarded over long range connectivity such as over a cellular infrastructure; the propagation messages are always broadcasted with shorter range connections until some node matches the request propagation with some response. Whenever a propagation message matching occurs, the user who placed the request message receives the response message via long range connectivity. Moreover, the users can pre-determine the type of queries to handle in some specific geographic context, called Bazaar. So, any person close or distant to a Bazaar can send requests through the cellular infrastructureto other users in a specific geographic (Bazaar). Information is spread around to users in a specific geographic location, but it does not benefit from the number of meetings among nodes in a specific geographic area as in SOLAR.
PeopleNet differs from the proposal defined in as it does not concern itself with any mobility pattern, similarity of mobility and the ability of nodes to learn about their own mobility. PeopleNet relies on the innovative idea of using the widely deployed cellular infrastructure and Bluetooth devices to propagate information search. A second peculiar contribution involves the overlay routing according to the meta-information (i.e. Bazzar and message type) to choose what connection must be used to forward the messages. So, one could extract the importance from high level information that could be used in routing.
SPROUT presents a social mechanism to route messages in overlay networks such as Distributed Hash Tables (DHT). It is based on the use of the knowledge of a trust relationship among social nodes to choose what node must receive a message and to associate message priority. SPROUT presents possible trust function according to the number of hops in a social overlay (the distance dij between social nodes n1 and nj). The relation in (46) is one of its trust functions used to choose the next hop, where f is a static probability for two nodes to be trusted friends. The reader may note that the probability of two nodes being friends is limited by the value r. For example, if f = 0.95 and r = 0.6, then the trust function assumes that the friends with high proximity of node n i are best friends (reflecting a high trust relationship) and consequently very likely to correctly route a message and when dij >8 the trust will be maintained as 0.6. The objective of this work is to reduce the number of several network attack types that may drop the packets or forward the data to any different node other than the correct destination. For instance, in a DHT structure, a malicious node may exchange messages in order to disseminate unwanted information. SPROUT locates the trustiest friend of a given node that has a closer identifier, but not greater than, a key value until finding the destination node for that given key. Should this fail, then the source node executes the traditional DHT process. Although this work has been implemented in overlays, it could also be applied at the network layer. A node should evaluate the trust relationship of its neighbor instead of choosing the shortest path, which may be an unsafe path in terms of optimization and integrity. Further, trust identification remains a hard undertaking.
On the other hand, the trust relationship is not immune to attacks completely. Malicious nodes may convince a small number of honest ones trough the creation of several and false identities to increase their influence and credibility in the network. Looking at this problem SybilGuard and Syb-ilLimit map users and nodes, separating the network in two groups: a honest region with nodes with only one identity and a Sybil region populated by malicious nodes with more than a single identity. They also established that the number of links between these regions (called attack edges) is independent of the number of malicious identities. Moreover, if a trust route contains only nodes into the honest region then all the routes that cross the same node or edge will converge. Therefore, one may observe that there are several works applying social approaches in order to improve information propagation in overlay networks, security and optimization.
Propagation in social overlay and underlay is very similar, but there is a little difference. Both the optimization and security are interlaced in the context of social networks. This may be the case when messages are dropped or wrongly forwarded. A number of security enhancements have been suggested in to improve underlay security. User's devices are used to route data in DTN scenarios, increasing the likelihood to disseminate worms and viruses, as their users are often inexperienced with regard to such security threats. Moreover, networks with high-degree nodes tend to connect to other high-degree node networks (the famous often move in the same circles) and are therefore more likely to be subject to epidemics. Indeed a single infected high degree node will quickly infect other high-degree ones. On the other hand, networks where high-degree nodes tend to connect to low-degree nodes show the opposite behavior; a single infected high-degree node will not spread an epidemic very far.
No comments:
Post a Comment