Technological advances, decreasing production costs and increasing capabilities have made sensor networks suitable for many application fields such as various scientific and commercial applications including warehouse management, battlefield surveillance and environmental monitoring [1–4]. Thanks to the advanced technology, over time sensor networks have been adopted in an all-around industry. Gathering data to be aware of any states by using those sensors is achieved by modeling it as a distributed database where sensor readings are collected and processed using queries [5–7].

Especially, sensor node components of sensor networks obtain the state information from sensor device parts on those nodes and store those data. Accordingly, each sensor node in a sensor network is regarded as a distributed database system generating a data stream and has been studied as a sensor database [7]. In query processing of sensor networks, the join operation costs much in sensor networks for correlating sensor readings like distributed database environments [8]. Therefore, many researchers have studied reducing the cost in sensor environment [1].

In a sensor network environment, a query is issued for retrieving and gathering the real time state information. The form of a well-used query in a sensor network is using an SQL-like declarative language [8]. The collected data in a sensor network can be seen as one distributed relation over the sensor nodes, called the sensor relation. The query operations are also served restrictively because of the limitation of the environments. Further, most previous solutions either assume that nodes have sufficient memory to buffer the partition of the join relations assigned to them for processing, or that the amount of memory available at each node is known in advance and the assigned data partitions can be set accordingly [1]. Under these assumptions, it is hard to apply assumptions to real life.

Therefore, we consider the communication cost aspect and we propose an Incremental Join Algorithm (IJA) as an in-network join strategy which is an efficient join processing in sensor networks and minimizes communication cost. The IJA strategy is capable of reducing communication cost and utilizing data by gathering real-time state information from sensors which is one of sensor network features. In sensor network environments, it is hard to send all data stored at each node to server located in the center as we consider in the assumptions. Therefore it needs to be filtered whether data are sent or not. The problem in the previous studies is that the results processed in the previous steps would be ignored as a query happens. However, as a different point between the earlier studies and this paper, this algorithm for processing query uses the previous result and just sends the operations needed to be joined to the final node. Therefore assume previous join results are stored in temporary repository to process efficiently. To evaluate the performance, we compare the IJA strategy to the conventional algorithms including synopsis strategy. The remainder of the paper is structured as follows. In the next section, we describe typical join strategies in sensor networks including synopsis to compare. In Section 3, we introduce and explain an incremental join algorithm. And we analyze the performance and compare IJA to typical join strategy including synopsis algorithm. In Section 5, we conclude with future works.

In sensor network, sensor nodes are formed by hundreds and hundreds of fixed nodes. Consequently, the value obtained periodically from sensor nodes is lack of expressing all the information about event or entity. It needs a join operation for that problem. The data stored at each sensor node forms a kind of table over all nodes, denoted R. To process a join query, we first have to decide which join queries are used. In this paper, we consider binary equi-join (BEJ). A BEJ query for sensor networks is defined as follows:

Definition 1 Given two sensor tables  R ( A 1 , A 2 ,     ... ,     A n )   and  S ( B 1 , B 2 ,     ... ,     B n ) ,   a binary equi-join (BEJ)   is R ∞ Ai = Bj S           ( i  ∈     { 1 , 2 ,     .... ,   n } , j  ∈     { 1 , 2 ,     .... ,   m } )where A_i and B_j are two attributes of R and S respectively, which have the same domain.

Consider a sensor network covering a road network from [7]. Each sensor node can detect the IDs of vehicles in its close vicinity, and record the timestamps at which the vehicles are detected. Suppose N_R and N_S represent two sets of sensor nodes located at two regions of a road segment, Region1 and Region2, respectively. To gather the necessary data for determining the speeds of vehicles traveling between the two regions, the following join query can be expressed:

SELECT R.autoID, R.time, S.time

To evaluate the above query, sensor readings from Region1 and Region2 need to be collected and joined on the autoID attribute. Typical join strategies of sensor networks are classified into Naïve join, Sequential join and Centroid join according to the join location and shown in Figure 1 [9].

One of general join problems is a heavy communication cost to be transferred into among nodes and a lower join selectivity of query regarding overhead in communication. For instance, given there are two tables, R and S, pairs which is not participated in joining operation between R and S tables can be sent to another region F to join with another table. In the Naïve join algorithm in Figure 1(a), sensor nodes around the sink node in region F are join nodes N_F selected. Although the cost of routing join results to the sink node can be minimized, the each whole table in region R and S is routed to the sink.

In the Sequential join algorithm in Figure 1(b), it is minimized by routing the join results to the sink after performing the local join R_i ∞ S where R_i is the local table stored at node n_i in region R and S is the table in region S. In this algorithm, the problem is also that the whole table in region S is delivered to the nodes in region R. That makes communication cost be high.

As compared with the earlier algorithms, the Centroid algorithm in Figure 1(c) could also deliver the tables in each region into region F. In spite of that delivery, the nodes which are close to each region R and S in distance are selected to be joined and it could minimize the communication cost. However, the tables in each region are also needed to be routed into the join region F.

To solve or reduce the problems above, the synopsis strategy join was suggested. After reducing the number of pairs in R and S tables using synopsis to remove the rest pairs not to participate join operation, SNJ sends the pairs to join with others. The means of synopsis is an abstract of a table to process join operation. In addition, the size of the synopsis table is smaller than original table size. Therefore, each sensor creates its synopsis [9]. Synopsis strategy consists of 3 steps. First is synopsis join step. The second step is notification and third is final join operation.

The first join operation of synopsis is as follows: each node, n_i ∈ N_R, stores the local table R_i which is one of local tables consisting of table R. Also each node n_i creates local synopsis S_i (R_i) by extracting join attributes A_j, and counting the frequency of the same value in the table. Synopsis join region N_L is selected to get a final join candidate pairs from joining table R and S synopsis. The synopsis join nodes after receiving synopsis from N_R and N_S synopsis process synopsis join operations.

The second step, a notification of synopsis join strategy, notifies final join candidate pairs to the N_R and N_S nodes. For this, synopsis join node n1 stores sensor ID of local synopsis originated. In the third step, each node of N_R or N_S notified from synopsis join nodes n₁ sends join attribute v to final join node n_f. The final join node n_f joins with R_v ∞ S_v, and then sends the results to query sink node.

However, although the synopsis algorithm has contributed to reduce the communication cost, the algorithm only excludes duplicated data and a lot of data could be delivered as far as the table has various attribute values not to be duplicated. Therefore, we need to process as a unit of query and suggest an incremental join algorithm.

In this section, we propose an incremental join algorithm (IJA) to gather and process real-time state information which is one of the sensor network features. First, we describe general environment components including terms in next section [7,11] and the algorithm later.

In this chapter, we evaluate the performance of IJA and compare it with typical join strategies such as naïve join, sequential join and centroid join including synopsis strategy. The experiment is mainly measured by the total number of messages incurred for each join strategy because join processing in sensor network is a complex operation due to the distributed nature of the processing and the limited memory at nodes. Other comparisons for performance evaluation and experiments will be included in our future work.

Sensor networks have been adopted in various scientific and commercial applications. Gathering data from sensors is achieved by modeling it as a distributed database where sensor readings are collected and processed using queries. Sensor nodes are generally highly constrained, in particular regarding their energy and memory resources. While simple queries such as SELECT and AGGREGATE queries in wireless sensor networks have been addressed in the literature, the processing of join queries in sensor networks remain to be investigated. Previous approaches have either assumed that the join processing nodes have sufficient memory to buffer the subset of the join relations assigned to them, or that the amount of available memory at nodes is known in advance.

Therefore, in this paper including these assumptions, we describe an Incremental Join Algorithm (IJA) in Sensor Networks to reduce the overhead caused by moving a join pair to the final join node or minimize the communication cost that is the main consumer of the battery when processing the distributed queries in sensor network environments. To evaluate the experiments, we compare the IJA with the typical algorithms. including the synopsis algorithm which is a representative strategy in sensor network to process queries. We also show the result of comparisons. In case of under join selectivity 0.01, typical join algorithms, such as naïve, sequence and centroid join, perform worse than synopsis and IJA algorithms.

Despite having better synopsis performance compared to a typical join algorithms, IJA performs better than the synopsis algorithm in conditions with above 60% synopsis. As future work, we will vary and study the parameters such as network density, node memory capacity and synopsis size, including communication cost.