| Summary: | 博士 === 國立臺灣科技大學 === 資訊工程系 === 95 === Abstract
With the rapid development of wireless communication technology, mobile computing becomes possible. People with battery-limited mobile devices can access various kinds of information from anywhere at anytime. However, existing wireless services are limited by the constraints of mobile environments such as narrow bandwidth, frequent disconnections, and limited battery capacity. Data broadcast is an attractive approach in such context. Disseminating data through a broadcast channel allows simultaneous access by an unlimited number of mobile clients and thus allows efficient usage of scarce bandwidth.
Sometimes the broadcast bandwidth is composed of multiple channels, where channels are allocated on discontinuous spectrums and can not be combined to a shared channel. The use of multiple channels allows better fault tolerance, configurability, and scalability. In this dissertation, we consider two issues arises in case multiple broadcast channels are present: i) how should the broadcast data be allocated to various channels, and ii) how should mobile clients surf the channels to obtain the data of interest.
Access efficiency and energy conservation are two main performance issues in wireless broadcast system. Access efficiency concerns how fast a request is responded, and energy conservation concerns how to conserve a client’s battery power when it accesses the desired data. Efficient data accesses to multi-channel broadcast programs are motivated by the desire to satisfy client requirement efficiently with as little consumption of energy as possible.
Cyclically broadcasting data over a channel is a basic scheduling approach called flat broadcast. When multiple channels are available, a data allocation scheme is needed to assign data to channels. Some channels may be reserved for those few frequently requested items, while the infrequently accessed bulk of data are allocated on other channels, this kind of allocation scheme called skewed allocation. In view of data access skew (the access probabilities of data items are usually different), in Chapter 3, we study the problem of broadcasting N data items over K channels assuming skewed data allocation and flat broadcast per channel, with the object of minimizing the average expected delay of the clients. The previously known DP algorithm is a straightforward dynamic programming implementation to partition data items and achieves an optimal solution with cost O(KN2). We propose a restricted dynamic programming (RDP) approach to achieve a similar performance to DP while with a much cheaper cost O(NlogK). Simulation results show that, the hit rate obtained by RDP is higher than 90% and it also outperforms similar work 200%. By using the curve fitting technique, we further refine RDP and develop an O(NlogNlogK) PKR algorithm; simulation results show that the solution is optimal
In Chapter 4, we study fast access to data that are broadcast on multiple channels. If a client has not a priori knowledge of how data are allocated to them, in the worst case a client may have to scan all the channels before finding the desired data. We consider the access protocol and structure for making effective use of multiple channels. We propose adding index information to a skewed data allocation on a well-known index channel to help clients surf on data channels. Our solution consists of i) organizing data by popularity, which is done by PKR data allocation scheme; ii) organizing data by key, which is done by supplementing an index search tree for each data channel. Then nodes of all index trees are multiplexed onto a well-known index channel. Our method organizes data and index into a popularity hierarchy, thus the popular data and their indices are broadcast more frequently than less popular data and their indices. Simulation results show that the reduction of access time is on average about 48% over previous work with little tuning time overhead.
|
|---|
