Bounding Transport Complexity for Content Dissemination in Online Social Networks

Abstract—In this paper, we address the issue on measuring the transport difficulty for content dissemination in online social networks (OSNs). We define a new metric, called transport load, to measure the load imposed by the OSN on the carrier communication network. It involves two key factors: data arrival process at users and transport distance of messages. Furthermore, we define the fundamental limit of transport load as the transport complexity, i.e., the minimum required transport load for an OSN over a given carrier communication network. To model the content dissemination in OSNs, we formulate the geographical distribution feature of dissemination sessions in OSNs by introducing a four-layered system model. It consists of physical layer, social layer, content layer, and session layer. We dig the mutual relevances among these four layers as the preconditions for bounding the transport complexity. Specifically, taking into account the scale-free property of users’ degree centrality in OSNs, we introduce the Zipf’s distribution to describe the friendship degree of users. Furthermore, in order to model the interest-driven session patterns in real-world OSNs, we define a new session pattern, called Social-InterestCast. Combining with the users’ behaviour in real-life OSNs and the experimental results of our evaluation, we make a reasonable assumption that: For the session with a source node, say v, the number of destinations chosen from all friends of v follows a Zipf’s distribution. Based on our system model, we derive the lower bound on transport complexity for the content dissemination in a large-scale OSN under this common probability distribution in the social sciences. To the best of our knowledge, this is the first work to measure the transport difficulty of data dissemination applications in OSNs by comprehensively handling the diversity of social session patterns in terms of the scale-free properties of both users’ degree centrality and dissemination behavior.