3.1 Overview of the HRDD
We assume that a wireless sensor network has the following properties:
• After randomly deployment, sensor nodes remain stationary at their initial locations.
• The sensor nodes are homogeneous and location unaware.
• The sensor nodes can make decisions without global information.
• There are multiple mobile sinks which generate queries in the WSNs.
• The sensor nodes communicate with sinks by delivering data across multiple hops.
The hierarchical structure can reduce communication overhead and data redundancy in the sensor networks. We use clustering technique to build a hierarchical structure that each mobile sink can easily maintain its data dissemination path. Figure 3.1 illustrates the fundamental concept of a cluster hierarchy, and the number in the figure expresses the ID of sensor node. All sensor nodes organize themselves into low-level clusters via a CH election process [3] [4] [5] [9] [18]. The low-level clusters, in turn, organize themselves into high-level clusters.
15
Figure 3.1: Example of Hierarchical Role-based Data Dissemination (HRDD)
In HCDD, when a sink registers its location information, sink’s high-level CH will broadcast regis-tration message to other high-level clusters through border clusters as shown in Figure 3.2. Since one of border clusters may overlap the neighboring high-level clusters of another border cluster, high-level CH may get reduplicate sink’s registration messages. It will generate redundant messages and make entire sensor network consume much energy. In addition, as a sink issues a query, each high-level CHs local query forwarding to their low-level CHs. In the example shown in Figure 3.3, since the sensing data are stored to low-level CH, each high-level CHs must forward query to them to get the relevant data. However there may not be any relevant data in the low-level CH, it will consume much energy caused these unnecessary messages. Since the registration of mobile sink’s location increases much energy consumption and lots of messages, and data delivery from source to sink may induce unneces-sary query messages. So, we assign two roles, Indexing Agents and Gateway Agents to some nodes.
Indexing Agents, which are the border nodes in high-level cluster (Figure 3.5. the red nodes), should store the event messages of neighboring low-level clusters, and Gateway Agents, which are the border clusters belong to high-level cluster (Figure 3.4. the green nodes), should allocate broadcasting path
High-level CH
Ordinary Node Low-level CH
Sink Location Registration
Figure 3.2: Sink Location Registration in HCDD
High-level CH
Ordinary Node Low-level CH
Query Data
Figure 3.3: Query Data in HCDD
to other high-level clusters. Once a source detects an event, the sensing data are stored to its low-level
High-level CH
Indexing Agent Gateway Agent
Ordinary Node Low-level CH
Sink Location Registration
Figure 3.4: Sink Location Registration in HRDD
High-level CH
Indexing Agent Gateway Agent
Ordinary Node Low-level CH
Query Data
Figure 3.5: Query Data in HRDD
CH and then an event message is forwarded to the Indexing Agent which the low-level CH belongs to.
When a sink issues a query, it must broadcast its registration message to all high-level clusters through Gateway Agent for source data forwarding. In this way, sink can easily get the relevant data by querying Indexing Agents, instead of querying to all low-level CHs, and sink lightly sends its registration mes-sage direct through Gateway Agent to other high-level clusters, not through the border clusters belong to high-level cluster. Hence, sink queries by Indexing Agents could avoid unnecessarily transferring the query messages between source and sink for data lookup, and sink registration through Gateway Agent could decrease energy consumption of broadcasting and number of flooding messages.
There are five procedures in HRDD. First, in Cluster Construction, all nodes are grouped into multilevel clusters, and each cluster will elect a node as the CH. Second, in the Selection of Indexing Agent and Gateway Agent, each high-level CH has to select a set of nodes as Indexing Agents or Gateway Agents by Agent Selecting Algorithm. Third, in Event Detection, the sensing data is stored to its low-level CH and an event message send to Indexing Agent for events and queries. Fourth, in Sink Location Registration, each sink registers its location information to all high-level CHs through Gateway Agent, Finally, in Query Data Forwarding, when an Indexing Agent has relevant data which a sink queries, the data will be forwarded to the sink by high-level CHs and low-level CHs along the reverse path. The detail of each procedure will be presented in the following subsections.
3.2 Cluster Construction
There are a number of clustering techniques been proposed in the literature [3] [4] [5] [9] [18]. In the Linked Cluster Algorithm (LCA) [4], a node becomes the CH if it has the highest identity among all nodes within one hop of itself or among all nodes within one hop of one of its neighbors. The Distributed Clustering Algorithm (DCA) [5] uses generic application-dependent weights associated with nodes to elect CHs. The Weighted Clustering Algorithm (WCA) [9] elects a node as a CH if it has the highest weight among its one-hop neighbors based on a combination of node characteristics that include node degree, transmission power, mobility and batter power of node considered. The Voting-based Clustering Algorithm (VCA) [18] votes for their neighbors to elect suitable CHs Voting-based on the local information of each node, such as residual energy and node degree.
All of the above algorithms [4] [5] [9] [18] generate one-hop clusters, require synchronized clocks and have O(n) complexity, where n is the number of nodes in the network. This makes them suitable only for networks with a small number of nodes. The Max-Min D-cluster algorithm proposed in [3]
generates D hops clusters based on node-ids with a run-time of O(D) rounds and only O(D) messages per node, where the value of D is selected by users. Hence, in our work, we group the sensors by Max-Min D-cluster algorithm, which formed by node IDs along the wireless links and can provide load-balanced clustering for extending the lifetime of the WSNs. Max-Min D-Cluster Formation Algorithm consists of a distributed CHs election algorithm, guaranteeing that no node is more than D hops away from its CH. The algorithm has four logical stages: The first stage uses D rounds to propagate the
largest node ID which nodes heard from other neighbor nodes. The second stage uses D rounds to propagate the smallest node ID which nodes heard from other neighbor nodes. In the third stage, each node selects its CH based on the information received in the first and second stages. Finally, each non-CH node communicates with its non-CH to join the non-CH’s cluster in the fourth stage. After the Max-Min D-cluster algorithm, CHs form a virtual backbone and may be used to route packets for nodes in their cluster. These CHs are called low-level CHs. Then, Max-Min D-cluster algorithm is performed on the low-level CHs to form high-level clusters.
3.3 The Selection of Indexing Agent and Gateway Agent
In HRDD, Indexing Agents are the rendezvous area to save the event messages of neighboring low-level clusters, and Gateway Agents allocate broadcasting path to other high-level clusters. After Cluster Con-struction, the high-level CHs are aware of local information, i.e. the information of low-level clusters and the neighboring high-level clusters. They use these local information to select Indexing Agents and Gateway Agents by Agent Selecting Algorithm. The main idea of Agent Selecting Algorithm is that selecting a set of nodes plays two roles. These nodes take place in the border nodes of high-level cluster or the border clusters of high-level cluster. In this way, when a sink issues a query, it is relaxed to communicate to other high-level clusters through Gateway Agents and is easily to query the rele-vant data by Indexing Agents, rather than using heavily broadcasting to other high-level clusters and searching hardly for all the low-level clusters of high-level CH.
There are two input parameters for Agent Selecting Algorithm, one is agent candidates, another is agent candidates’ neighboring clusters. For Indexing Agents, agent candidates are the border nodes in the high-level CH’s low-level cluster (in Figure 3.6. high-level CH 73’s Indexing Agent candidates are node 1, 7, and 35), and neighboring clusters are the neighbor low-level cluster of high-level CH’s low-level cluster (in Figure 3.6. high-level CH 73’s neighboring low-level clusters are cluster 65, 85, and 100). For Gateway Agents, agent candidates are the border clusters belong to high-level cluster (in Figure 3.6. high-level CH 73’s Gateway Agent candidates are cluster 65, 85, and 100), and neigh-boring clusters are the high-level CH’s neighbor high-level clusters (in Figure 3.6. high-level CH 73’s neighboring high-level clusters are cluster 89, 92, and 99).
The Agent Selecting Algorithm consists of the following two phases: