Mobile sinks are proposed to save sensor energy spent for multihop communication in transferring data to a base station (sink) in Wireless Sensor Networks. Due to relative low speed of mobile sinks, these approaches are mostly suitable for delay-tolerant applications. In this paper, we study the design of a query scheduling algorithm for query-based data gathering applications using mobile sinks. However, these kinds of applications are sensitive to delays due to specified query deadlines. Thus, the proposed scheduling algorithm aims to minimize the number of missed deadlines while keeping the level of energy consumption at the minimum.
A Wireless Sensor Network (WSN) can be defined as a network of sensor nodes deployed to monitor a field with wireless communication capability and base stations (sinks) to gather information from sensors for uploading to a remote central. Usually sensor nodes are powered by unrechargeable batteries. When a sensor depletes its battery, it becomes nonfunctional which can affect the connectivity and correctness of WSN. Therefore, energy consumption is a crucial factor affecting the life time of a WSN.
Various energy conservation approaches have been proposed and implemented so far to maximize network lifetime as surveyed in [
In this paper, we present a novel way of using MS for a near real time application, namely, query-based data gathering with deadlines, by trading off delay in response with energy consumption in Multihop communication. In this class of applications, location-based queries are submitted to WSN, and responses should be collected before the specified deadline expires. For this reason, we design a query scheduling algorithm to exploit MS deterministic mobility for saving energy in communication and to exploit speed of Multihop communication for minimizing delay caused by slow MS motion, whenever any of them is feasible. Thus, our algorithm balances the system throughput and energy consumption by optimizing the number of hops and duration of response time. The algorithm is very simple yet successful and applicable to the situations where either controlling MS moves is not possible (e.g., geographic conditions) or feasible (e.g., attached to a public transport vehicle).
The paper is organized as follows: related work is presented in Section
The possible ways of initiating data transfer from sensors to a sink can be categorized into four classes: event-driven, time-driven, query-based, and hybrid [
For any data collection approach, sensor readings should be transferred to some sink which can be classified as static sink (SS) and mobile sink (MS) according to mobility. Sink mobility affects the communication pattern between sensor and sink [
MS mobility patterns can be listed as controlled, random, or deterministic [
In this work, we focus on query-based data collection in WSN using mobile sink with deterministic mobility pattern. Likewise, in [
In another similar work, authors model a WSN in which sensors store collected data in their finite memory [
There are other query based data collection approaches which mainly focus on designing a high level interface for query—response interactions between application and sensors. Unfortunately, these approaches do not work on details of underlying network topology, communication requirements, and energy consumption issues [
The
We assume that a WSN has been deployed for monitoring some environmental changes such as heat and mobility, as shown in Figure
Assumed WSN model.
Any sensor node which is onehop away from the route can directly communicate with passing by MS. These sensors are called
Whenever the remote central initiates a query, MS attempts to create a schedule such that the communication cost (energy consumption) for forwarding messages would be minimum and response would be uploaded before the given deadline (the algorithm can be run by remote central or MS. In this work we assume that MS runs the algorithm whenever it receives a query). A schedule is composed of
While preparing schedule, MS keeps moving and releases queries to related QRS when it is in one-hop proximity of QRS. This trip, from current MS location when it received the query to QRS location, is named
Query and response message content.
As MS is passing by RCS, it attempts to collect the response message from the sensor. The trip from QRS to RCS is called
To meet the deadline in the WSN model given above, we need to ensure that MS reaches RCS before deadline and the response message reaches RCS before MS. Similarly to minimize the energy consumption in transferring query and response packages, we have to select QRS and RCS from GS such that they have the shortest path (minimum number of hops away) from the target sensor. However, these two conditions could not be always satisfied. Thus, the main scheduling problem is to find QRS and RCS such that MS can obtain the response before the deadline and routing costs the minimum energy consumption. Below we provide the details of Deadline-Aware Energy-Efficient Query Scheduling (DES).
DES first attempts to construct
As summarized in Algorithms
(1) Select QRS and RCS (2) Calculate (3) (4) (5) run MECS algorithm (6) (7) (8) (9) (10) (11) (12) (13) (14) run OECS algorithm (15) (16) (17)
and
(1) Select ARCS s.t. Dis(ARCS, TS) is minimum and updated (2) Select AQRS s.t. updated (3) (4) (5) (6) Select ARCS and AQRS s.t. Dis(ARCS, AQRS) is maximum, updated (7) (8) (9) (10) (11) (12)
The details of creating schedules are given below.
If deadline permits to create the shortest routing path between QRS to RCS via TS, the energy saving would be maximized since the number of hops are expected to be minimum as seen in Figure
Least Energy Consuming Schedule when MS moves in the direction of TS (a) and when it moves in the other direction (b).
After deciding QRS and RCS, LECS checks several parameters to see if the query deadline can be met with the current schedule. The duration of Query Release Trip Time (
After calculating all these parameters, LECS first checks if MS has enough time to reach RCS location before the deadline. If deadline allows, MS should ensure the time needed for query forwarding and processing, and response forwarding would be less than trip time to RCS. In some cases response message would be late to arrive at RCS, and as a second chance, we might allow MS to move up to RE and come back to RCS. When response can reach RCS before the extended response collect trip time, we should check if deadline does not still expire (one can suggest that even in an extended trip time it is not enough, and deadline does not expire; therefore, MS can execute another tour on the path back to RCS once more. Since the communication speed is much more than the MS speed, we ignore this case).
According to all these conditions we either schedule QRS and RCS successfully, or we call other scheduling algorithms to calculate alternative paths. If LECS algorithm fails to create a shortest routing path due to late arrival of response message to RCS, it calls OECS to select an alternative QRS (AQRS) such that MS can release query earlier and WSN would have more time to route the response message to RCS. On the other hand when LECS algorithm fails because of deadline expiration before MS finishes collect trip, MECS algorithm is called for choosing alternative QRS (AQRS) and alternative RCS (ARCS) such that MS and the response message would meet at ARCS before the deadline. Consider
As seen in Algorithm
Optimum energy consuming schedule when MS moves in the direction of TS (a) and when it moves in the other direction (b).
Using RCS selected by LECS algorithm, OECS algorithm calculates alternative QRS (AQRS) location between
As discussed above, LECS attempts to use minimum energy by constructing shortest routing paths for query and response message delivery, whereas OECS consumes least energy for response messages but more energy for delivering query to gain time. Whenever these two algorithms fail to create a feasible solution, as a last resort, they call MECS algorithm. The routing path constructed by MECS algorithm costs more energy to gain time by attempting to select an alternative QRS (AQRS) as well as an alternative RCS (ARCS) in the hope that response message can be reachable by MS before the deadline.
As shown in Figure
Most Energy Consuming Schedule when MS moves in the direction of TS (a) and when it moves in the other direction (b).
Considering deadline if such AQRS location is not available, MECS algorithm recalculates AQRS and ARCS locations such that the distance between these two locations would be the largest. We hope that while MS moves from AQRS to ARCS, the query and reply message can be forwarded up to ARCS before deadline. If MECS algorithm again fails to find such AQRS and ARCS locations, the query will be rejected.
This section presents the evaluation of Energy-Efficient Deadline-Aware Scheduling with respect to several performance metrics and two other scheduling methods.
Table
Simulation parameters and default values.
Parameters | Definition | Default setting |
---|---|---|
|
Width of monitored field | 1000 m |
|
Height of monitored field | 500 m |
|
Number of sensors | 1771 |
T | Sensor topology | Grid |
ST | Simulation time | 54000 s (15 h) |
RT | MS route location | Horizontal center |
RR | Radio range | 50 m |
DR | Data transfer rate | 256 Kb/s |
EC | Data transfer cost | 50 nJ/bit |
BP | Initial battery power | 0.01 J |
QP | Query processing time | 0.1 s |
PC | Query processing cost | 100 nJ |
MS | MS speed | 40 km/h |
QS | Query size | 32 Byte |
RS | Response size | 256 Byte |
QA | Query arrival rate | Exp. (mean = 30 s) |
DL | Deadline | min. 17 s/max. 74 s |
Performance metrics are as follows.
For a given query generation rate distribution, any scheduling algorithm is subject to similar number of queries. However, the number of submitted query will depend on different parameters such as the current network connections, sink position, battery power of sensors, and scheduling algorithm. Ideally all generated queries should be submitted to WSN. On the other hand, to save sensor energy, scheduling algorithms can reject submitting queries whose response would not arrive on time. If the implemented scheduling algorithm is successful in selecting these kinds of queries, WSN network life would extend. Otherwise, if it fails in prediction, then system throughput will be decreased considerably. Contrary to the prediction of late response arrival, a scheduling algorithm could submit all the generated queries hoping that the responses would arrive on time.
Thus, to compare success of our algorithm,
For the default values of parameters given in Section
Results of various performance metrics related with query and response numbers.
Results of Successful Query Ratio for different query deadline values.
Results of Network Life Time performance metric for different battery levels.
In Figure
Figure
Figure
As a second parameter to calculating energy spending of the data collection methods, we present the Average Energy Consumption Per Submitted Query in Figure
Results of Average Energy Consumption Per Submitted Query for different query deadline values.
For a larger value of deadline (74 sec.), we obtained the following results presented in Figure
Results of various performance metrics related with query and response numbers for a larger deadline.
Query Arrival Rate (QA) follows exponential distribution with a mean value 30 as default. To simulate different query loads on WSN, we change the QA mean value to 15 and 45. The results of Successful Query Ration and Average Energy Consumption Per Submitted Query performance metrics for different query load values are presented in Figures
Results of Successful Query Ratio for different query loads.
Results of Average Energy Consumption Per Submitted Query for different query loads.
Since MS/IS just submits any generated query immediately expecting to collect the response whenever MS/IS passes by the RCS, MS/IS does not react to changes in QA when the Successful Query Ration metric is observed. However, SS produces poor success query ratio when QA is high, because, for a fixed simulation time, with a higher query arrival rate, SS needs to submit more queries which leads to early battery depletion of surrounding sensors. As a result, all forthcoming queries are to be rejected. With lower QA, SS can submit more of them via its surrounding sensors before their batteries get empty. MS/DES presents better results for Successful Query Ration metric with lower QA. However, MS/DES outperforms the other approaches for all different query arrival rates.
For Average Energy Consumption Per Submitted Query performance metric, MS/IS consumes different amount of energy, whereas MS/DES and SS consume similar amount of energy when a different level of QA is applied. Since MS/IS immediately submits incoming queries to the nearest sensor and expects responses to arrive to the route via the shortest path, the difference in Energy Consumption Per Submitted Query occurs only in the query path. When there are more queries in a higher QA, the differences are increased.
On the other hand, since SS is located at the center of WSN and query and response paths are the same mostly, Average Energy Consumption Per Submitted Query is not affected by the different query load. For MS, scheduling of queries is based on the algorithm which aims to minimize the energy consumption for each query, MS/DES can sustain the same level of energy spending successfully. Thus, even MS/DES is not fixed at a location, it could accomplish similar level of success in Average Energy Consumption Per Submitted Query performance metric as SS.
This paper introduced a scheduling algorithm for the location-based queries in WSN with a mobile sink following a deterministic mobility pattern. In WSN, Multihop communication pattern is used to disseminate the queries and the responses. The queries have associated with deadlines. The proposed scheduling algorithm aims at maximizing the number of successful queries and reducing the sensors’ energy expenditure due to Multihop communication by exploiting deterministic sink mobility. For this reason, before submitting queries, the scheduling algorithm selects the release and collect sensors such that two important performance requirements can be met: the energy required to forward data packages should be minimum, and the response arrival time should not exceed the specified deadline.
We also simulated two data collection methods for the sake of comparison, namely, SS and MS/IS. We conducted extensive simulation tests, and the obtained results show that our scheduling algorithm can attain more successful queries with less amount of energy even when query load and deadline change.
As a future work, we would like to extend and adapt the algorithm to different mobility models other than linear route. We plan to apply some heuristics such as Ant Colony Optimization techniques to decide minimum energy consuming paths while MS decides its own route.