International Journal of Computer Information Systems,Vol. 4, No. 1, 2012
randomly and then move along the designated paths using ashortest-path computation. Group-based mobility modelsintroduce dependency among the mobile nodes. Johanssonet al. propose a Disaster Area scenario, in which individualgroups consisting of rescue agents intercommunicate witheach other . Hu and Johnson use a pursue model, againsuggested by Sanchez, in which nodes follow a group leaderby trying to intercept it . Hong et al. develop a ReferencePoint Group Mobility model, which is a generalization of the pursue model . In this model, each node belongs to a
group with a logical center, and a node’s velocity is defined
as the sum of the velocity of the center its own randomvelocity. By adjusting the movement of the logical center,this model can be used to produce various real-life group-based scenarios, such as disaster management, a conventioncenter, etc.Finally, several mobility frameworks have been developedto characterize a wide variety of movements. The MobilityVector model  uses a pair of vectors to model smoothchanges in direction and speed. They show how variousmobility scenarios can be generated from this basic model,including location-dependent movement, targeting, andgroup mobility.Topic-based addressing is an abstraction of numericnetwork addressing schemes. With the content-basedsubscriptions used in SIENA and Gryphon, deliverydepends only on message content, extending the capabilityof event notification with more expressive subscriptionfilters . Common topic-based systems arrange topics inhierarchies, but a topic cannot have several super topics.Type-based subscription provides a natural approach to thisif the language offers multiple sub-typing, thus avoidingexplicit message classification through topics. This workswell with typed languages, but it is complex to deploy thisdegree of serialization of objects. Moreover, mobileapplications may not have the concept of objects or typing.Thus, the combination of hierarchical topics and high speedcontent filtering could be a more flexible approach formobile applications. There are efforts to build content-basedsubscription with distributed hash tables by automaticallyorganizing the content into several topics. Research is alsoongoing to structure complex content-based data models and reflection-based filters . XRoute  proposesan approach for content based routing of XML data in mesh-based overlay networks.We are currently working on efficient distribution of eventmatching tasks over the network and brokers by establishing
the concept of ”approximate matching”. The dynamic
construction of event dissemination trees to route eventsfrom publishers to all interested subscribers is the mostchallenging task to support content-based subscription indistributed environments. JEDI proposes variations forevent routing among its networked event servers, includingthe flooding and match-first approaches. With thehierarchical approach, event servers are organized in a tree.This approach may lead to a large routing table at the treeroot. Routing strategies in SIENA use two classes of algorithm: advertisement forwarding and subscriptionforwarding. They prune the propagation tree by propagatingonly those paths that have not been covered by previousrequests.
2.1 Publish/Subscribe System
Currently available publish-subscribe middleware differalong several dimensions among which the most relevantare the expressiveness of the subscription language, thearchitecture of the dispatcher, and the forwarding strategy,. The expressiveness of the subscription languagedraws a line between subject-based systems, wheresubscriptions identify only classes of messages belonging toa given channel or subject, and content-based ones, wheresubscriptions contain expressions (called predicates) thatallow sophisticated matching on the message content. Weguess that the typical scenarios of pervasive computing,where a potential large number of components need tointeract very flexibly, justify the latter choice with respect tothe more conservative, less expressive, and less scalablesolution of subject-based filtering. To apply content basedpublish/subscribe in large scale networks, most advancedmiddleware adopt a distributed dispatcher, where a set of brokers are interconnected in an overlay dispatchingnetwork and cooperatively route subscriptions and messagessent by components are attached to them. Middleware thatexploit a distributed dispatcher can be further classifiedaccording to the interconnection topology of brokers and thestrategy exploited for message dissemination. The simplestapproach is message forwarding in which brokers areconnected to form a un-rooted tree. Publishers sendmessages to their associated broker, which forward them toall the other brokers by following the tree topology.Moreover, each broker keeps track of the subscriptionscoming from the software components directly attached to itin a local subscription table, which are used to determine thecomponents, that should receive incoming messages. Thissolution inevitably results in high overhead as all messagesare sent to all brokers, regardless if an attached componenthas subscribed.
Multicasting has traditionally been used as transport for
messaging systems. However today’s multicast schemes are
not scalable to support large groups. Thus, severalApplication-Level Multicast Routing Protocols (ALMRPs)have been designed. Most ALMRPs use tree based routingfor logarithmic scaling with respect to receiver numbers: asthe node connectivity changes, the tree structure changesaccordingly. Narada  and many subsequent designs can
JanuaryPage 38 of 63ISSN 2229 5208