IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 4, JULY 2006 1347 Exploiting Locality of Demand to Improve the Performance of Wireless Data Broadcasting Petros Nicopolitidis, Georgios I. Papadimitriou, Senior Member, IEEE, and Andreas S. Pomportsis Abstract—With the increasing popularity of wireless networks and mobile computing, data broadcasting has emerged as an effi- cient way of delivering data to mobile clients having a high degree of commonality in their demand patterns. In many applications, clients are grouped into several groups, each one located in a different region, with the members of each group having similar demands. In fixed-bit-rate wireless broadcast systems, transmis- sion power is set at such a level that guarantees the necessary level of received energy per bit for all clients in the service area so that they can operate under a predefined bit error rate level. However, as in wireless cellular environments, the path loss of wireless signals is typically inverse to the fourth power of the trans- mitter/receiver distance, there exists an increasing redundancy in the level of received energy per bit for decreasing distances from the server’s antenna. This paper proposes a mechanism that exploits locality of demand in order to increase the performance of wireless data dissemination systems. Specifically, it trades the received energy per bit redundancy at distances smaller than the radius of the service area for an increased bit rate for transmission of items demanded by clients at such distances. This results in an increased transmission speed for many items. The bit rate for an item transmission is dynamically determined from the distance between the server’s antenna to the group of clients that demand this item. Knowledge of clients’ positions is conveyed to the server via a simple feedback from the clients. Simulation results that reveal significant performance improvement over fixed-bit-rate broadcasting in environments characterized by locality of client demands are presented. Index Terms—Adaptive data broadcasting, asymmetric wireless environments, learning automata, locality of demand, variable bit rate. I. I NTRODUCTION D ATA broadcasting has emerged as an efficient means for the dissemination of information over asymmetric wire- less networks [1]. Examples of data broadcasting applications are traffic information, weather information, and news distrib- ution systems. In such applications, client needs for data items are usually overlapping. Consequently, broadcasting stands to be an efficient solution, as the broadcast of a single information item will likely satisfy a (possibly large) number of client re- quests. Moreover, in many applications, such as weather infor- mation and news distribution, the locations of clients determine their demands. Communications asymmetry is due to a number of facts, such as asymmetry in equipment (e.g., lack of client transmission Manuscript received June 4, 2004; revised November 12, 2004, April 20, 2005, October 4, 2005, November 9, 2005, and November 11, 2005. The review of this paper was coordinated by Dr. W. Zhuang. The authors are with the Department of Informatics, Aristotle University of Thessaloniki, Thessaloniki 54124, Greece (e-mail: gp@csd.auth.gr). Digital Object Identifier 10.1109/TVT.2006.877464 capability and client power limitations), asymmetry in the net- work system (e.g., small uplink/downlink bandwidth ratio), and application asymmetry (e.g., traffic pattern of client–server applications). The goal pursued in most of the proposed data-delivery approaches is twofold: 1) determination of an efficient sequence (broadcast program) for the transmission of the server’s data items in a way that the average response time (overall mean ac- cess time among the clients) is minimized and 2) management and operation of client local memory (cache) so that a client’s performance degradation is reduced when mismatches occur between the client’s demand pattern and the server’s program. This paper focuses on the minimization of response time under dynamic and location-dependent client demand patterns. So far, three major approaches have appeared for the server’s broadcast program. 1) In the pull-based approach (e.g., [2]), the server broad- casts information after explicit requests made by the mobile clients via the uplink channel. This approach is able to adapt to dynamic client demand patterns; however, it is inefficient from the point of view of scalability. This is because when the client population becomes too large, the client requests will either collide with each other or saturate the server. 2) In the push-based approach (e.g., [3]–[5]), the server is assumed to have an a priori estimate of the demand per information item and makes item broadcasts according to these estimates. Push systems provide high scalability and client hardware simplicity since clients do not need to include data packet transmission capability. However, push systems are unable to operate efficiently in envi- ronments with dynamic demand patterns. Nevertheless, with minimal changes to client and server hardware, Nicopolitidis et al. [6] extends the applicability of the push approach to environments characterized by a priori unknown and dynamic client demands and presents re- sults that reveal efficient operation in such environments. 3) Hybrid approaches (e.g., [14]) try to combine the benefits of the pure-push and pure-pull approaches. However, they need to carefully strike a balance between push and pull and manage a number of additional issues (determination and dynamic selection of bandwidth available for push and pull, selection of items to be pushed and those to be pulled, etc.). Information dissemination applications can be characterized by locality of client demands. A possible example of this case could be the case of a museum possessing the necessary 0018-9545/$20.00 © 2006 IEEE