Evaluation and comparison of various indexing schemes in single-channel broadcast communication environment

Jiaofei Zhong, Weili Wu, Xiaofeng Gao, Yan Shi, Xiaodong Yue
2013 Knowledge and Information Systems  
Wireless Data Broadcasting is a newly developed data dissemination method for spreading public information to a tremendous number of mobile subscribers. Access Latency and Tuning Time are two main criteria to evaluate the performance of such system. With the help of indexing technology, clients can reduce tuning time significantly by searching indices first and turning to doze mode during waiting period. Different indexing schemes perform differently, so we can hardly compare the efficiency of
more » ... ifferent indexing schemes. In this paper, we redesigned several most popular indexing schemes for data broadcasting systems, i.e., distributed index, exponential index, hash table, and Huffman tree index. We created a unified communication model and constructed a novel evaluation strategy by using the probability theory to formulate the performance of each scheme theoretically and then conducted simulations to compare their performance by numerical experiments. This is the first work to provide scalable communication environment and accurate evaluation strategies. Our communication model can easily be modified to meet specific requirements. Our 123 376 J. Zhong et al. comparison model can be used by the service providers to evaluate other indexing schemes to choose the best one for their systems. Introduction Wireless Data Broadcasting becomes more and more popular in recent years because of its scalability and flexibility to disseminate public information to a mass number of mobile subscribers with common interests, since it can satisfy all pending requests of the same data in one single response. In a typical data broadcasting system, during some time periods, a group of data items (named as a program) are broadcasted periodically as RF radio signals by a base station within a certain area. Clients located in the valid region can access broadcasting channel, search for the required data item, wait until the data item appears, and then download it. In practice, a number of real-world applications utilize data broadcasting techniques, where mobile clients have common interest on a certain group of data. For instance, locationbased information such as local attractions, news, traffic, and weather can be broadcasted to visitors who travel to some place for the first time. In addition, wireless broadcasting service and devices by Ambient (www.ambientdevices.com), for example the 7-day weather forecaster, sports devices, as well as the Weather and Information Center Refrigerator by LG Electronics, demonstrate the industry's interest in wireless data broadcasting. Since majority mobile devices have limited battery power and constraint lifetime, access latency and tuning time are two main criteria to evaluate the performance of a data broadcasting system. Considering a process from the moment when a client initiates a query to the moment it finishes downloading the data item, access latency denotes the whole time interval of this process, while tuning time denotes the sum of time when a client keeps "active" during the process. According to the architectural enhancements, each mobile device has two modes: active mode and doze mode. It can operate in active mode and stay idle in doze mode. Usually, the energy consumed in active mode is approximately 20-30 times higher than that in doze mode. Therefore, access latency evaluates the query response time of a system and tuning time evaluates the energy efficiency. Indexing technologies have been introduced to reduce tuning time for a data broadcasting system. An index is a specific data structure containing the location information of data items. Due to the nature of data broadcasting, indices in data broadcasting system store the "time offset" of target data items. Once a client gets this time offset, it is aware of when the target data item will be broadcasted on the channel. Then, it can turn off to doze mode to save energy and tune in again right before the data item appears. Different indexing technologies have different searching efficiencies. Moreover, if we insert indices into data items, then the whole size of a program will increase, resulting in longer access latency. Therefore, when discussing about an indexing scheme, researchers will always consider the balance between tuning time and access latency. A lot of traditional disk-based indexing techniques have been modified to fit the requirement of data broadcasting systems, e.g., distributed index [10], Huffman tree [11], spatial index [12], hash table [29] , exponential index [27], signature tree [31], but they are constructed under different environments, which brings difficulties to compare their performance. Moreover, the same indexing technique may perform quite differently under distinguished situations. Therefore, it is desirable to construct a unified evaluation strategy to analyze the
doi:10.1007/s10115-013-0643-9 fatcat:hgeuxmzwenhxlhibetk6kxdkku