An efficient indexing framework for data dissemination in wireless sensor networks

Abstract

Wireless Sensor Networks grow rapidly due to the sensing and scanning capabilities of nodes. The sensors are capable to move randomly and send periodic reports, hence they must be processed efficiently. This requirement increases the indexing burden, in particular, indexing the mobile sensors’ data. The grid-based indexing techniques in random mobile sensors suffer from high dependability on packet attributes and the necessity of registering the current cluster location of each sensor and assign its destination periodically. Therefore, this work proposes a framework called Dynamic-Coalition to efficiently improve the indexing packets of random mobile sensors. The proposed framework alleviates the disseminated packets by constructing dynamic-coalitions using relevancy-measure. The dynamic-coalition framework is superior in terms of average routing overhead and packet delivery ratio. It can significantly outperform R-tree, Decomposition-tree, and Grid-based DBSCAN with Cluster Forest in terms of index building time overhead, the number of traversed nodes, and space cost overhead.

Introduction

Wireless Sensor Networks (WSNs) attracted researchers in recent years and have a significant potential in new applications consisting of a large number of sensor nodes such as environmental monitoring . A WSN is a structure that includes many stable or mobile sensors deployed geographically. The deployed sensors transfer their packets to targeted destinations periodically. These sensors can be mobile in certain network applications, and their locations change through time. The need to store, process, and present mobile packets in a seamless, efficient, and interpretable form requires an efficient indexing technique. Random mobile sensors, namely, the type of mobility that we emphasize in this work, exacerbate the difficulty of the indexing process.

Indexing difficulty is due to the characteristics of reported packets in random mobile sensors. In a random environment, packets have the following characteristics: volume, velocity, scalability, and variability. Given these characteristics, the reported packets in random mobile sensors are considered a type of big data. Velocity refers to the high speed of creating and updating sensor packets. It increases the number of indexing operations (update, delete) to handle the speed of reported packets. For scalability, packets are generated continuously from different sources (sensors), which increase index building time and space cost. For variability, packets are transferred from different mobile sensors that dynamically change with location and time. Therefore, the mentioned random mobile sensors’ characteristics exacerbate the need to promote the indexing technique. The promoted indexing technique should be able to deal with these characteristics.

Grid-based techniques are proposed to efficiently index the packets of random mobile WSNs. The fundamental principle underlying grid-based techniques is clustering packets in accordance with partitioning network areas. Grid-based techniques construct the index structure on the basis of the included mobile sensors in each grid. Index construction depends on the sensors’ spatial and temporal attributes. Packet attributes are represented as the time period and the spatial location of the packets. For example, a network area is partitioned accordaning to sensor density in each grid, such as in DCSSC [1], DBSCAN [2], G-DBSCAN [3], and VDBSCAN [4]. A network area is also partitioned by the sensor and packet-based divisions, such as in R-tree [5], Frequent Updates in R-trees (FUR-tree) [6], R-trees with Update Memos (RUM-tree) [7], and R-tree with dynamic non-uniform grid and sub-R-tree [8]. However, the refered methods are ineffective for indexing random mobile sensors because of repeatedly changing sensor locations.

Grid-based DBSCAN with Cluster Forest (GDCF) [9] and Decomposition-tree (D-tree) [10] can be good alternative solutions for the aforementioned challenges. GDCF partitions a space layout into grids by utilizing the HyperGrid Bitmap (HGB) structure. HGB enables indexing grids that include objects and neglects empty ones. It provides neighbor grid merging using the union-find algorithm and cluster forest. D-tree attempts to index multidimensional mobile data to alleviate the high space cost in its inner nodes. It applies a hierarchical index structure that compromises B-tree, and R-tree. The D-tree structure can reduce the time taken for accessing memory by constructing its structure without inner nodes.

Although the refered tree-based index techniques have high popularity in traditional indexing, they fail to efficiently manage the packets of random mobile sensors. This failure is caused by many factors. First, the nature of the reported packets is considered a type of big data. Second, current indexing techniques depend on the partitioning of network areas to build the index structure. Third, they fail to detect unpredictable behaviors, such as random mobility, in which the type of mobility must be considered. These factors increase the indexing overhead because they provide considerable updating (read-write) operations, especially in random environments. The following measures can be performed to avoid the challenges brought by the previous factors. First, the index-tree performance is improved by mitigating the disseminated packets throughout the destinations. Second, the random mobile sensors’ packets not only refer to big data but also cover the packet attributes in which grid-based indexing is based on.

For further explanation, Fig. 1 represents the environment of random mobile sensors, labeled as part a. The environment consists of deployed sensors that move randomly, labeled as S0, S1, S2, and S3. The sensors transfer packets to random mobile gateways, labeled as gw1, gw2, and gw3. The details of the transferring packets of each sensor are represented in part b of Fig. 1. S0 transfers packets at times B and C to gateway 1 and at times A and C to gateways 2 and 3, respectively. The details of the received packets in each gateway are shown in Fig. 2, part a. A problem occurs when the destination receives packets from different times and locations but belong to the same source, i.e., S0 in gateway 1 and S1 in gateway 2 as represented in part b Fig. 2. This condition is considered a source of the frequent updating problem because it negatively affects the indexing performance when it accumulates on the final destinations for indexing.

To this end, the significance of this work is to diminish the effects of the aforementioned factors, which lead to the following drawbacks. First, they exacerbate the dependability on the multi-attributes of packets to build the index structure. Second, they result in the burden of periodically registering the information of cluster location and the need to allocate the destination for each sensor. To overcome these drawbacks, we propose a framework called Dynamic-Coalition, which constructs blocks called dynamic-coalitions, using a relevancy-measure. The proposed framework replaces grid-based techniques with constructed coalitions to alleviate the effects of disseminated packets. The framework positively influences the efficiency of indexing random mobile sensors.

The remainder of this study is organized as follows. A literature review is provided in Section 2. Section 3 presents the Dynamic-Coalition framework. The index-tree structure is discussed in Section 4. Performance analysis and results are explained in Section 5. Finally, Section 6 provides the conclusion and future work.