WIRELESS COMMUNICATIONS AND MOBILE COMPUTING Wirel. Commun. Mob. Comput. (2007) Published online in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/wcm.549
Reward oriented packet filtering algorithm for wireless sensor networks Lei Shu1,*,y ,z , Jie Yang2, Lin Zhang1, Hui Xu2, Xiaoling Wu2 and Manfred Hauswirth1 1 2
Digital Enterprise Research Institute, National University of Ireland, Galway, Ireland Department of Computer Engineering, Kyung Hee University, Seoul, Korea
Summary Packet filtering used in query process is an important approach to reduce energy consumption in sensor networks. Interest is used as a constraint to filter uninterested data when users query information. Within interested data some are more important because they have more valuable information than that of others. We hope to process these more important data first. In this paper, we firstly define reward to denote the importance level of data, and then we present a two tiers buffer model as the platform for our ETRI packet filtering algorithm. We especially present two packet filtering algorithms for real time and non-real time transmissions. Based on simulation result, we find out that both packet filtering algorithms can substantially improve the quality of information and reduce energy consumption. Copyright # 2007 John Wiley & Sons, Ltd.
KEY WORDS:
reward; packet filtering algorithm; sensor network; two tiers buffer; ETRI algorithm; query process
1. Introduction Conventional research that is proposed to be energy efficient, such as dynamic voltage scaling , has been fully utilized in all kinds of embedded systems. However, researches on communication subsystems which have not been fully analyzed can still provide plentiful enhancement in terms of reducing wireless communication energy consumption. Since wireless communication is the main consumer of battery’s energy, dynamic modulation scaling (DMS) has been proposed in Reference [1] to schedule packet transmission. In addition to the task scheduling in real time operating systems which are based on the CPU’s processing, extending the concept of DVS into the commu-
nication subsystems to provide a better packet scheduling can substantially reduce the energy consumption. The key idea is to let radio transmit packets with a lower transmission rate to reduce the energy consumption while still meeting all deadlines. Some similar researches [2,3] also follow this approach by applying lazy scheduling algorithm. These researches are focusing on minimizing energy consumption of a set of packets by delaying the finish of transmission till the deadline. One common drawback of these researches is that they only consider the packets that already exist inside the buffer, but do not provide the threshold or constraint to filter and reduce the incoming packets. Another energy-efficient research trend is presented in paper [4–6]. Rusu et al. first time consider energy,
*Correspondence to: Lei Shu, Digital Enterprise Research Institute, National University of Ireland, Galway, Ireland. y E-mail:
[email protected] z This work was partially accomplished when Lei Shu was studying as a graduate student at the Ubiquitous Computing Laboratory in Kyung Hee University. Copyright # 2007 John Wiley & Sons, Ltd.
L. SHU ET AL.
time, and reward constraints simultaneously while reward denotes the important level of tasks. Naturally, among a set of tasks of real time applications, some of them are more valuable than the others. They believe that instead of processing several unimportant tasks that just consume a small amount of energy, it is more meaningful to process one valuable task that will consume more energy, especially in some overload systems. Therefore in this energy, time, and reward (ETR) scheduling algorithm, whenever a new task is processed, it must have the highest ratio (reward value/energy consumption of this task) among the waiting tasks, and its energy consumption should not exceed the remaining energy. Moreover, by considering these three constraints simultaneously, system designers can determine the most important components of their system, or emphasize a subset of the system over another in a dynamic fashion. An example of this flexibility is that Zhang et al. [7] extend this ETR algorithm for packet scheduling and present three different transmission algorithms in paper: (1) throughput maximization under time/energy constraints; (2) value maximization under time/ energy constraints; (3) energy minimization under time/value constraints. Packet filtering is also an important approach to reduce the energy consumption as well as provide high quality information to users in sensor networks. If we can reduce the receiving and transmitting times by refusing and discarding the unnecessary data, then we can substantially reduce the energy consumption. Generally, a huge amount of data can be created by a large sensor network. However, in most of the time only the data of some sensor nodes that related to the end user’s purpose is really valuable. In other words, only those data in which end user is interested are useful. In References [8–10], data-centric approach is proposed for energy-efficient data routing, gathering, and aggregation in sensor networks. Interest is introduced as one constraint which is used to filter and reduce the unnecessary data as well as improve the information quality. In these researches authors simply consider that all these packets have the same important level, but actually among these interested packets, some of them may be more important than others. For example, users are interested in the data of several sensor nodes which are used to monitor one object. The data created by the sensor nodes which are close to the observed object have more valuable information than the data created by the sensor nodes which are far from this object. Therefore, if we can introduce the reward into these interested packets, we Copyright # 2007 John Wiley & Sons, Ltd.
are able to select out and process the most important and valuable packet first. In this paper, we present a two tiers buffer model (buffer of sensor networks and buffer in sensor node) to be the platform for our ETRI packet filtering algorithm. We especially present two packet filtering algorithms for real time and non-real time transmissions. Within these algorithms each packet has four parameters as follows: (1) energy consumption of the packet; (2) processing time of the packet; (3) important level of the packet; and (4) interest level of the packet. By using ETRI packet filtering algorithms, we can achieve two major contributions: (1) using interest constraint as the threshold to filter the uninterested incoming packets to reduce the energy consumption; (2) using reward constraint to choose the high quality information and minimize the queried packet number to minimize the energy consumption but still satisfy the minimum information requirement. The remainder of the paper is structured as follows. In the next section, we describe some related work. In Section 3 we describe the concept of reward. Section 4 presents the query APIs of ETRI. In Section 5 the system model is presented. In Section 6 we describe ETRI packet filtering algorithm. We present the simulation and discussion results in Section 7. Finally, this paper is concluded in Section 8. 2.
Related Work
Heterogeneous sensor networks are envisioned to consist of large numbers of devices, each capable of some limited computation, communication, and sensing, operating in an unattended mode. One unifying view is to treat them as distributed databases. The simplest mechanism to obtain information from this kind of database is to use queries for data within the network. However, most of these devices are battery operated, which highly constrains their life-span, and it is often not possible to replace the power source of thousands of sensors. So how to querying with the limited energy resources on the nodes is a key challenge in these unattended networks. Researchers have noted the benefits of a query processor-like interface to sensor networks and the need for sensitivity to limited power and computational resources [11–15]. Prior systems, however, tend to view query processing in sensor networks simply as a power-constrained version of traditional query processing: given some set of data, they strive to process that data as energyefficiently as possible. Typical strategies include minimizing expensive communication by applying Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
REWARD ORIENTED PACKET FILTERING ALGORITHM
aggregation and filtering operations inside the sensor network. In Reference [16], the authors present a sensor information networking architecture called SINA, which facilitates querying, monitoring, and tasking of sensor networks. To support querying within sensor networks, they design a data structure to be kept inside the sensor nodes based on the spreadsheet paradigm. In the spreadsheet paradigm, each sensor node in a sensor networks maintains a logical datasheet containing a set of cells. A node creates a new cell when it receives a request from other nodes, particularly from the user or its cluster head. By defining the semantic of a cell to specifying scope of the query, the information can be organized and accessed according to specific application needs, and also the number of the packets need to be sent can be reduced, thus the energy consumption will be reduced. However, there exist a tradeoff between the energy cost to run SINA on each sensor node and the energy reduced by using SINA. And also no simulation results are provided to demonstrate SINA’s efficiency. Madden et al. discussed the design of an acquisitional query processor (ACQP) for data collection in sensor network in Reference [17]. They provide a query processor-like interface to sensor networks and use acquisitional techniques to reduce power consumption. Their query languages for ACQP focus on issues related to when and how often samples are acquired. To choose a query plan that will yield the lowest overall power consumption, the query is divided into three steps: creation of query, dissemination of query, and execution of query. Optimizations are made at each step. Optimizer focuses on ordering joins, selections, and sampling on individual nodes. Our work has some similarities to techniques proposed in References [18–20]. The authors introduced a new real-time communication architecture (RAP) and also a new packet scheduling policy called velocity monotonic scheduling (VMS). VMS assigns the priority of a packet based on its requested velocity. VMS improves the deadline miss ratios of sensor networks by giving higher priority to packets with higher requested velocities, which also reflects the local urgency. RAP provides the following query/ event service APIs. A query is issued a query for a
query results will be automatically sent from an area to the issuer of the query in every period. This differs from our work in two aspects: one is that the costmodel is different in the two scenarios—RAP is primarily concerned with reducing the end-to-end deadline miss ratio while we are primarily concerned with minimizing energy consumption and maximizing the querying quality; the second one is that RAP intends to maximize the number of packets meeting their end-to-end deadlines without considering their value (reward, importance level), and in our model, we take reward an important constraint to deal with the queries.
3.
Understand Reward
Reward is defined to be the importance level of the data collected by sensors. In the case of different wireless sensor networks, it can be specified to various formats. We provide one example to make readers clearly understand the concept of ‘reward’. We consider a set of heterogeneous sensor nodes which are distributed over an area to detect an emitter or illuminated target at a specific location, as shown in Figure 1. The scenario may correspond to monitoring the presence of people, vehicles, or military targets using radar-like sensors that emanate specific signals into the region of interest [21–23], or passive sensors that passively detect emission levels associated with the object of interest. We assume the target may appear at a random position in a region, but the position of each sensor is fixed. At the sensing stage, we assume that a number of sensor nodes are involved in the signal detection, for example, N. Here a simple geometric
query{attribute_list, area, timing_constraints, querier_loc} list of attributes in an area. A query has timing constraints. If a period is specified for a command, Copyright # 2007 John Wiley & Sons, Ltd.
Fig. 1. Detecting an emitter or illuminated target. Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
L. SHU ET AL.
path loss model [24], and the path loss is proportional to 1/r, where is the path loss exponent, which is an environment-dependent constant typically between 2 and 4. For sensor node i, we use yi to denote the received signal. Thus, at sensor node i, the observations under two different hypotheses are given by ( yi ¼
=2 Rti
ni
þ ni
Hi H0
ð1Þ
Where H1 denotes the target-present hypothesis and H0 is the null hypothesis; ni is zero-mean, complex Gaussian variance 2; is a scalar defined ffiffiffiffiffiffiffiffiffiffiffiwith ffi pnoise by ¼ etr =2 ; and Rti denotes the distance between target and sensor node i. Obviously, sensor nodes near the target will have better observations, which means that the importance levels of these observations are much higher than those of the observations collected by further sensors. Thus, here we can consider reward to be the observations from sensor nodes about the target. Consider another example: lots of sensors are deployed in some area with different densities. For the EventFound case [25,26], take noise into account, the data collected by sensors having higher densities will be more reliable. So, here the reward is changed to be the density of the sensors around the interested area in the network.
4.
ETRI Query/Event Service APIs
Interested area specifies the scope of the query, the area from which data are needed by the users. Interested level specifies how the users are interested in the queried attributes. System value is defined as the sum of selected packets’ reward. Timing constraints can be period, deadline, and so on. If a period is specified for a command, query results will be sent from the interested area to the issuer of query periodically. The querier_loc is the location of the base station that sends out the query. A query is send to every node in the interested area specified in the API, and the results will first be sent back to the cluster head, then the cluster head will use the algorithm we will introduce in following sections to decide the packets to be sent back to the base station of which the location is also provided by the API.
5. 5.1.
System Model Working Scenario
Assume that we have one cluster in the heterogeneous sensor networks that is deployed as Figure 2 shows to monitor three different parts. Sensor nodes in area A, B, and C are used to monitor three different targets {a, b, c}. After issuing query and sensing operations, only the data collected by the sensor nodes which are located in area A and B can be accepted by the cluster head. Data from the area C will be rejected by the cluster head, because the end user is not interested in them. However among these two locations A and B,
Applications may submit queries or register for events through a set of query/event service APIs. The APIs provides a high-level abstraction to applications by hiding the specific location and status of each individual node. These APIs allow applications to specify the timing constraints as well as other constraints of queries. ETRI provides the following query/event service APIs. query{attribute_list, interested_area, interestd_level, system_value, timing_constraints, querier_loc} A query is issued for a list of attributes in an interested area with the maximum system value (reward). Attributes refer to the data collected by different types of sensors, such as temperature sensors, humidity sensors, wind sensors, rain sensors, etc. Copyright # 2007 John Wiley & Sons, Ltd.
Fig. 2. Sensor nodes in areas A, B, and C are used to monitor three different targets {a, b, c}. Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
REWARD ORIENTED PACKET FILTERING ALGORITHM
this user is more interested in the object A’s information. Therefore, whenever two packets are sent simultaneously by area A and B respectively, the cluster head will accept the packet from area A first. In this sense, we can use the value-based approach to describe the interest level of packets. If we look inside area A, we can find that the data sensed by the sensor nodes which are relatively closer to the target have the higher valuable information. Thus, we consider these sensor nodes’ data more important than others’. Then, whenever the cluster head receives the packets from sensor nodes, it will receive the most valuable packet among several interested packets first. Similar with the interest, we also can use reward (value-based) to describe the important level of packets. We provide an example for readers to understand the query operation based on ETRI query/event service APIs. We suppose that an end user sends a query to know the information about targets {a, b} in area A and B simultaneously. Collected information should arrive at base station within end-to-end deadlines of 4 and 5 s, respectively. The query includes the location of base station so that query results can be sent back. In this paper, we assume the location of base station is fixed. The general query sent to cluster head is as follows: query{observation_a and observation_b, area_A and area_B, Interested_Level_A ¼ 6, Interested_Level_B ¼ 5, system_value, Deadline_A ¼ 4, Deadline_B ¼ 5, Base_station ¼ (250, 350)} After receiving this general query, cluster head creates two sub-queries for areas A and B, as follows. Referring to Equation (1), we use the Rai and Rbj denoting the observations encapsulated in packets created by sensor nodes i and j in areas A and B, respectively. Here the observation value is considered as reward. Thus, the system_value is the sum of observation values of selected packets created by different sensor nodes. Interest value is used to denote the interest level of packet. A packet with a larger interest value means that this packet is more interested by users. In this example, users are more interested in target a’s information since the interest value is set as 6 which is larger than target b’s interest value 5. sub_query_A{observation_a, area_A, Interested_Level_A ¼ 6, system_value ¼ Ra1 þ Ra2 þ þ Rai, Copyright # 2007 John Wiley & Sons, Ltd.
Deadline_A ¼ 4, Cluster_head ¼ (120, 150)} sub_query_B{observation_b, area_B, Interested_Level_A ¼ 5, system_value ¼ Rb1 þ Rb2 þ þ Rbj, Deadline_B ¼ 5, Cluster_head ¼ (120, 150)} 5.2.
Two Tiers Buffer Model
In terms of the cluster head, as the Figure 3 shows, many unprocessed packets are still physically existing in different sensor nodes and waiting for the processing of cluster head. Therefore, in sensor network, except the cluster head, all the other sensor nodes which are going to send packets to the cluster head can logically be considered as a buffer, since all of these packets are waiting for the processing of cluster head. We regard this buffer as the First Tier Buffer (FTB). Actually the FTB is a logical concept for cluster head. The Second Tier Buffer (STB) is the buffer that physically exists inside cluster head. Since many sensor nodes will send packets to cluster head, obviously, cluster head needs buffer to store these received packets. Therefore, we propose the two tiers buffer model for wireless sensor network. The two tiers buffer model can be used in LEACH [27] or other cluster based sensor networks [28–32]. It enables network designers to put diverse packet filtering algorithms in FTB as well as diverse packet scheduling algorithms in STB. By clearly separating these two tiers buffer, network designers can dynamically combine different packet filtering algorithms
Fig. 3. Two tiers buffer model. Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
L. SHU ET AL.
and packet scheduling algorithms for different working purposes. 5.3.
Missions of Both Buffers
In FTB, we have the following two options to filter packets: (1) Consider interest constraint only [8–10]: Sensor nodes can distinguish packets based on their content and will not accept those packets which they are not interested in. (2) Consider both interest and reward constraints: Similar with option (1), sensor nodes only accept their interested packets. However, among these interested packets, sensor nodes accept more important packets first. In STB, we use two different algorithms to transmit packets for real time and non-real time situations: (a) Consider non-real time transmission For non-real time transmission, how to use energy more efficiently is the major problem. We decide to use DMS [33] to reduce energy consumption during packets transmission. In DMS, system designers compromise the energy consumption and execution delay. Whenever sensor node processes the incoming packets, it always considers the energy as the first class resource, and slows down the processing speed in order to just meet the deadline. (b) Consider real time transmission We transmit accepted packets to base station by using VMS. As the algorithm that has been presented in Reference [18], which assigns the priority of a packet based on its requested velocity (velocity ¼ distance/deadline). VMS minimizes the deadline miss ratios of sensor networks by giving higher priority to packets with higher requested velocities, which also reflects the local urgency. VMS embodies with both the timing constraint and location constraint.
The four constraints of ETRI algorithm [34–36] are defined as follows: The energy constraint imposed by the total energy Emax available in the cluster head. The total energy consumed by the accepted packet should not exceed the available energy Emax. In other words, whenever the cluster head accepts one packet, the energy consumption Ex,y of this packet should not be larger than the remaining energy RE. The time constraint imposed by the global deadline D. The common deadline of this user’s data query is D. Each packet that is accepted and processed must finish before D. The interest constraint imposed by the interest value threshold IT. Each packet that is accepted and processed must satisfy the interest value threshold ITmin Ix;y ITmax . The reward constraint imposed by the value ratio Vx,y (Vx,y ¼ Rx,y/Ex,y) between reward value Rx,y and energy consumption of packet Ex,y. The larger Vx,y, the packet has, the more valuable the packet is. The ultimate goal of packet filtering algorithm is to select out a set of packets P ¼ P1 [ P2 [ PM among interested packets to maximize the system value which is defined as the sum of selected packets’ value ratio Vx,y to meet the system_value defined in the query/event service APIs. Therefore, the problem is to maximize
X
Vx;y system value
ð2Þ
x2A;y2P
subject to X
Ex;y Emax
ð3Þ
Tx;y D
ð4Þ
x2A;y2P
X x2A;y2P
6. 6.1.
ETRI Packet Filtering Algorithm Mathematical Model
We define the interested areas as A fA1 ; A2 . . . AM g. From each interested area Ax the cluster head can accept a subset of packets Px fPx;1 ; Px;2 ; . . . ; Px;N g. The processing time of the packet Px,y is denoted by Tx,y. Associated with each packet Px,y there is an interest value Ix,y and a reward value Rx,y. Copyright # 2007 John Wiley & Sons, Ltd.
ITmin Ix;y ITmax
ð5Þ
x2A
ð6Þ
A ¼ fA1 ; A2 ; . . . ; AM g
ð7Þ
Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
REWARD ORIENTED PACKET FILTERING ALGORITHM
y 2 Px
ð8Þ
Px ¼ f1; 2; . . . ; Ng
ð9Þ
Since P ¼ P1 [ P2 [ . . . [ PM , we can have the following equation as: X X X Vx;y ¼ VA1 ;y þ VA2 ;y x2A;y2P
A1 ;y2P1
þ þ
X
A2 ;y2P2
VAM ;y
ð10Þ
AM ;y2PM
From Equation (10), we can find that the real problem of ETRI packet filtering algorithm is to find out the minimum subset of Px f1; 2 . . . Ng to maximize the system value to system_value from each interested area Ax. Thus, the problem is changed to X Vx;y system value ð11Þ maximize Ax ;y2Px
subject to Ex;y RE X
Tx;y D
ð12Þ
ð13Þ
y2P
ITmin Ix;y ITmax
ð14Þ
x2A
ð15Þ
A ¼ fA1 ; A2 ; . . . ; AM g
ð16Þ
y 2 Px
ð17Þ
Px ¼ f1; 2; . . . ; Ng
ð18Þ
Inequality (12) guarantees that the time constraint is satisfied. Inequality (13) guarantees that only the interested packets are accepted, and inequality (14) guarantees that the energy budget is not exceeded. In order to solve the problem that is presented by Equations (11)–(18), we give the following steps for our ETRI algorithm. Copyright # 2007 John Wiley & Sons, Ltd.
6.2.
Steps of ETRI Algorithm
Before sending the real data of a packet to cluster head, sensor node can send its packet’s parameters to the cluster head by including them in a small packet, which just consumes very limited energy. We give a name to this kind of small packet as Parameter Packet (PP). There is a physical buffer that exists inside cluster head to store these PPs. After receiving these PPs, cluster head can decide which packet to be accepted and which packet should be discarded based on these sent parameters. In terms of this two tiers buffer model, basically, we can define our ETRI algorithm into the following steps: Step 1: Initialization. After receiving PP fPP1 ; PP2 ; . . . PPN g, we assume that tables exist inside the cluster head for storing parameters of every packet iði 2 PP: energy consumption Ex,y, processing time Tx,y, reward value Rx,y, and interest value Ix,y. For each PPi, there are energy consumption for checking CEi and a period of time for checking CTi. We also use two structure arrays, considered(i) and selected(i) of size N, to store the information for all received PPs. Initially, we start with an empty schedule (selected(i).status ¼ false) and no PP is considered (considered(i).status ¼ false). The set of selected PPs (initially empty) is defined as S ¼ {(i) | selected(i).status ¼ true}. After selecting the PPs, cluster head accepts packets that are corresponded to these selected PPs. Therefore, packet’s parameters can be expressed as considered(i).Ex,y, considered(i).Tx,y, considered(i).Rx,y, considered(i).Ix,y, selected(i).Ex,y, selected(i).Tx,y, selected(i).Rx,y, and selected(i).IP x,y. We define five variables: (1) checking energy ð i2PP CEi Þ is used to store the total energy Pconsumption for checked PPs; (2) checking time ð i2PP CTi Þ is used to store the total processing time for checked PPs; (3) processing P energy ð i 2 PP selectedðiÞ:Ex;y Þ is used to store the total energy consumption for processed packets; (4) P processing time ð i 2 PP selectedðiÞ:Tx;y Þ is used to store the total processing timePfor processed packets; and (5) system value summation ð i 2 PP selectedðiÞ:Rx;y Þ is used to store the total value for packets to be processed in STB. These five variables are all initialized to zero. Step 2: In FTB, we filter and accept packets based on the ETRI constraints. A packet that can be accepted should satisfy all the following criteria: This packet’s PP is not considered before (considered (i).status ¼ false). The current schedule is feasible (checking timeþ processing time) D. Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
L. SHU ET AL.
By accepting this packet to current schedule, the energy budget is not exceeded (checking energyþ processing energy þ considered(i).Ex,y Emax). This packet is intentionally queried by end user (ITmin considered(i).Ix,y ITmax). Among all the PPs that satisfy the above criteria, select the one that has the largest considered(i).Vx,y ¼ considered(i).Rx,y/considered(i).Ex,y. By accepting this packet to current schedule, the summation of the system value is just up to the system_value defined in the query/event service P APIs ð i 2 PP selectedðiÞ:Vx;y system valueÞ. After choosing the PP, cluster head can send Acknowledge back to accept new packet. In addition, for those packets which end user is not interested in, their corresponded sensor nodes will discard them. In this case, we refuse and discard the unnecessary data; consequently, we can reduce the energy consumption by reducing the data transmitting and receiving. The flowchart and pseudo code of ETRI algorithm are showed in Figures 4 and 5. Step 3: In STB, we transmit accepted packets to base station by using two options: (1) Using VMS for real time transmission; (2) Using DMS for non-real time transmission. Another aspect: Replace or drop a packet in the STB. A new packet is always accepted if possible. When receiving new PP from sensor node, if the STB is full, we can replace or drop a packet based on the following criteria: This packet’s PP is selected (selected(i).status ¼ trueÞ: Among all selected packet’s PPs, find out the one that has the smallest selected(i).Vx,y ¼ selected(i). Rx,y/selected(i).Ex,y. If this found one is not the new packet that is going to be accepted, we use this new packet to replace this found one, otherwise, we drop this new packet. In next section, we present the simulation environment as well as approach. The performance comparison with four algorithms is given based on simulation results. 7. 7.1.
Simulation and Discussion Simulation Environment
A sensor network can be modeled as a graph, where each vertex represents a sensor node and each edge Copyright # 2007 John Wiley & Sons, Ltd.
Fig. 4. Flowchart of ETRI packet filtering algorithm.
represents the edge between two nodes when they are within each other’s communication range. This network tracks the values of certain variables like temperature, humidity, etc. Application users submit their requests as queries and the sensor network transmits the requested data to the application. For the simulation work, we randomly deploy 11 different sensor nodes. And we randomly initialize these sensor nodes with: the total energy of sensor nodes (scope: from 111 to 888), the buffer size of sensor nodes (scope: from 6 to 9). Ten of these 11 sensor nodes are chosen to be the packet generators Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
REWARD ORIENTED PACKET FILTERING ALGORITHM
These 10 sensor nodes are organized into three groups based on their created packets’ interest values. The packets that have the interest values belong to {8, 9, 10} are considered as group A, the packets that have the interest values belong to {6, 7} are considered as group B, and the packets that have interest values belong to {3, 4, 5} are considered as group C. Suppose the cluster head just accepts the packets from area A and B, moreover, within these interested packets it accepts the packet that has the largest Vx,y ¼ Rx,y/Ex,y first. And we also design that this cluster head works in the STB by using the DMS and VMS respectively. In terms of energy consumption, we mainly consider the following two parts that have strong relationship with our proposed ETRI algorithm, which are processing energy{E(Returning ACK) þE(Receiving packet) þE(Processing) þ E (Broadcasting event) þ E (Listening) þ E (Accepting ACK) þ E(Sending packet)} and checking energy {E(Accepting event) þ E(Deciding)}. The checking energy is designed to be 0.3, which is 10 per cent of the minimum packet consumption 3; also the checking time is designed to be 0.3, which is 10 per cent of the minimum processing time 3. 7.2.
Algorithms for Comparison
We provide four different algorithms to run on the cluster head for comparison as follows: (1) Compared Algorithm 1 (CA 1):
Fig. 5. Pseudo code of ETRI packet filtering algorithm.
which randomly create these 10 different packets and send to the remaining one. The remaining one works as the cluster head. For this cluster head, we design five parameters: the total energy ¼ 666, the buffer size ¼ 6, the deadline ¼ 5, the system_value ¼ 10, and the interest threshold ¼ 5. The meaning of threshold is that we just accept the packets when their interest values are larger than 5. Packets from those areas are what the end users are interested in. In addition, we design 10 different packets that are randomly initialized with the following four parameters: energy consumption (scope: from 3 to 10), processing time (scope: from 3 to 10), reward value (scope: from 3 to 10), and interest value (scope: from 3 to 10). Copyright # 2007 John Wiley & Sons, Ltd.
(a) In FTB: No interest constraint and no reward constraint, (b) In STB: Minimizing the packet deadline miss ratio (VMS). The cluster head does not set any threshold to reduce the incoming packets, but just simply receives packets and relays them. Once it gets a packet, it will process this packet based on the velocity determined by time constraint and location constraint. (2) Compared Algorithm 2 (CA 2): (a) In FTB: Consider interest constraint, but no reward constraint, (b) In STB: Minimizing the packet deadline miss radio (VMS). The cluster head always accepts the packet that has the interest value larger than the interest threshold. Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
L. SHU ET AL.
Once it gets a packet, it will process this packet based on the velocity determined by time constraint and location constraint. (3) Compared Algorithm 3 (CA 3): (a) In FTB: No interest constraint and reward constraint, (b) In STB: Maximizing system lifetime (DMS). The cluster head does not set any threshold to reduce the incoming packets, but just simply receives packets and relays them. Once it gets a packet, it will always process this packet just meeting its deadline. (4) Compared Algorithm 4 (CA 4) [7]: (a) In FTB: Consider reward constraint, but no interest constraint, (b) In STB: Maximizing system lifetime (DMS). The cluster head always accepts the packet that has the largest value ratio among several checked packets. Once it gets a packet, it will always process this packet just meeting its deadline. 7.3.
Simulation Results and Comparisons
We use CA 1 and CA 2 to compare with ETRI (VMS) algorithm while running VMS in the STB. From Figure 6, we can find that algorithm CA 1 costs a lot of processing energy and our ETRI (VMS) algorithm costs only about half of that. The reason is that the cluster header just simply receives the packets and relays them without reducing any incoming packets, neither interest nor reward constraint is considered in Algorithm CA 1. Take a look at Figure 7, we find that the energy utilization ( ¼ processing energy/(checking energy þ processing energy)) of our ETRI (VMS) algo-
Fig. 6. Total processing energy. Copyright # 2007 John Wiley & Sons, Ltd.
Fig. 7. Energy utilization.
rithm is a little bit lower than the other two algorithms. Remember that we used both interest and reward constraints, which would definitely cost some checking energy, however, we still reduce the energy consumption of whole sensor networks. The saved energy comes from the normal sensor nodes but not from the cluster head. Same conclusion can also be drawn form Figure 8, by analyzing the discarding ratio (discarding ratio ¼ discarded packets/total created packets), we can see that the discarding ratio of our ETRI (VMS) is much higher than others. The lower discarding ratio the sensor nodes have, the more uninterested packets the sensor nodes send. Thus, the more unnecessary energy is consumed. In conclusion, by using the ETRI (VMS), the sensor nodes can reduce the unnecessary transmission of uninterested data to reduce the energy consumption. Consequently we get Figure 9 showing the total time consumption, even though we need more checking time, we reduce the total time consumption by processing only part of the packets. For this part, the packets have larger reward than that of the rest packets. As we presented in foregoing paragraph, we design the interest threshold to accept packets that have larger interest values, therefore, the desired average interest value should be larger than that of other algorithms. Figure 10 shows that the average interest value of ETRI (VMS) is larger than others, which means the ETRI (VMS) can exactly process the interested packets well. Figure 11 shows the comparison among three algorithms’ average reward values. In the algorithm CA 1, because we do not intentionally maximize the value ratio (Vx;y ¼ Rx;y =Ex;y ), as a result, the average reward value of CA 1 is relatively smaller than others. Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
REWARD ORIENTED PACKET FILTERING ALGORITHM
Fig. 8. Discarded packet ratio.
Fig. 9. Total time consumption.
Fig. 10. Average interest value.
Fig. 11. Average reward value. Copyright # 2007 John Wiley & Sons, Ltd.
Fig. 12. Lifetime of cluster head.
Compared with CA 2, even though we add the interest constraint to CA 2, still no reward constraint is considered, thus the average reward values of our ETRI (VMS) is the largest one. Similarly, we use CA 3 and CA 4 to compare with ETRI (DMS) algorithm while running DMS in the STB. From Figure 12, we can find that for a given amount of energy, by using the DMS technique, the lifetimes of three different algorithms are almost same. As Figure 13 shows, the total processing energy of ETRI (DMS) is less than that of others, which still
Fig. 13. Total processing energy. Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
L. SHU ET AL.
Fig. 14. Energy utilization.
because we add both interest and reward constraints at the same time. This consequently causes relatively low energy utilization of ETRI (DMS), as showed in Figure 14. Even though the energy utilization of ETRI (DMS) is relatively lower than others, by using ETRI (DMS) algorithm, we can still substantially reduce the energy consumption of whole sensor networks. The saved energy comes from the normal sensor nodes but not from the cluster head. By analyzing the Figure 15, we can find that the discarding ratio of ETRI (DMS) is much higher than others. Once again, by using the ETRI (DMS), the sensor nodes can reduce the unnecessary transmission of uninterested data to reduce the energy consumption. Figure 16 also shows that the average interest value of ETRI (DMS) is much larger than others, which also means the ETRI (DMS) can exactly process the user interested packets well. As Figure 17 shows, the algorithm CA 3 does not intentionally maximize the value ratio (Vx;y ¼ Rx;y =Ex;y ), as a result, the average reward value of CA 3 is smaller than others. Compared with CA 4,
Fig. 16. Average interest value.
Fig. 17. Average reward value.
even though we add the interest constraint in ETRI (DMS), however, the average reward values of two algorithms are still almost same. This means the ETRI (DMS) can inherit the original purpose of ETR scheduling well. Through all these simulation results, we demonstrate that our ETRI filtering algorithm can deal with the queries more efficiently and get more important information to solve the queries.
Fig. 15. Discarded packet ratio. Copyright # 2007 John Wiley & Sons, Ltd.
Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
REWARD ORIENTED PACKET FILTERING ALGORITHM
8. Conclusion Battery-operated heterogeneous sensor network should have a meaningful lifetime and provide high quality information to users. Packet filtering used in query process is an important approach to reduce energy consumption in sensor networks. Interest is used as a constraint to filter uninterested data when users query information. Within interested data some are more important because they have more valuable information than that of others. In this paper, after clearly defining the concept of reward and presenting a new two tiers buffer model, we propose a novel query model considering four constraints: Energy, Time, reward, and interest simultaneously, we can maximize the system value among the interested packet while satisfying the time and energy constraints. Based on our simulation results, we find out that our ETRI filtering algorithm can improve the quality of the information queried and also reduce the energy consumption. Acknowledgment This work was financially supported by Lion Project (grant no. SF1/02/CE1/131). References 1. Schurgers C, Raghunathan V, Srivastava MB. Modulation scaling for real-time energy aware packet scheduling. Global Communications Conference (GlobeCom’01), San Antonio, Texas 2001. 2. Prabhakar B, Biyikoglu EU, El Gamal A. Energy-efficient transmission over a wireless link via lazy packet scheduling. IEEE/ACM Transactions on Networking 2002; 10(4): 386–394. 3. Kompella RR, Snoeren AC. Practical lazy scheduling in sensor networks. ACM, SenSys’03, November, Los Angeles, California, USA, 2003. 4. Rusu C, Melhem R, Mosse D. Maximizing rewards for realtime applications with energy constraints. ACM Transacations in Embeded Computing Systems 2003; 2(4): 537–559. 5. Rusu C, Melhem R, Mosse D. Maximizing the system value while satisfying time and energy constraints. IBM Journal of Research and Development 2003; 47(5/6): 689–702. 6. Rusu C, Melhem R, Mosse D. Multi-version scheduling in rechargeable energy-aware real-time systems. Journal of Embedded Computing 2005; 1(2): 271–283. 7. Zhang F, Chanson ST. Throughput and value maximization in wireless packet scheduling under energy and time constraints. 24th IEEE International Real-Time Systems Symposium, 2003. 8. Krishnamachari B, Estrin D, Wicker S. Modeling data-centric routing in wireless sensor networks. Proceedings of 6th international workshop on modeling analysis and simulation of wireless and mobile systems, 2003. 9. Beaver J, Sharaf MA, Labrinidis A, Chrysanthis PK. Poweraware in-network query processing for sensor data. In Proceedings of the 2nd Hellenic Data Management Symposium, September 2003. Copyright # 2007 John Wiley & Sons, Ltd.
10. Tan HO, Korpeoglu I. Power efficient data gathering and aggregation in wireless sensor networks. SIGMOD/PODS, vol. 32, number 4, December 2003. 11. Intanagonwiwat C, Govindan R, Estrin D. Directed diffusion: a scalable and robust communication paradigm for sensor networks. In MobiCOM, Boston, MA, August 2000. 12. Madden S, Franklin MJ. Fjording the stream: an architechture for queries over streaming sensor data. In ICDE, 2002. 13. Madden S, Franklin MJ, Hellerstein JM, Hong W. TAG: A Tiny AGgregation Service for Ad-Hoc Sensor Networks. In OSDI 2002; 36(SI): 131–146. 14. Bonnet P, Gehrke J, Seshadri P. Towards sensor database systems. In Conference on Mobile Data Management, January 2001. 15. Wolfson O, Sistla AP, Xu B, Zhou J, Chamberlain S. DOMINO: databases for moving objects tracking. In ACM SIGMOD, Philadelphia, PA, June 1999. 16. Shen CC, Srisathapornphat C, Jaikaeo C. Sensor information networking architecture and applications. IEEE Personal Communications 2001; 8(4): 52–59. 17. Madden SR, Franklin MJ, Hellerstein JM, Hong W. The design of an acquisitional query processor for sensor networks. SIGMOD, June 2003, San Diego, CA. 18. Lu C, Blum BM, Abdelzaher TF, Stankovic JA, He T. RAP: a real-time communication architecture for large-scale wireless sensor networks. In IEEE RTAS 2002. 19. He T, Stankovic JA, Lu C, Abdelzaher T. SPEED: A stateless protocol for real-time communication in sensor networks. In Proceedings of the 23rd International Conference on Distributed Computing Systems (ICDCS-23), Providence, RI, USA, May 2003. 20. He T, Stankovic J, Lu C, Abdelzaher T. SPEED: A Real-Time Routing Protocol for Sensor Networks. University of Virginia Tech. Report CS-2002-09, March 2002. 21. Dutta PK, Arora AK, Bibyk SB. Towards radar-enabled sensor networks. In Proceedings of the 5th International Conference on Information Processing in Sensor Networks (IPSN) Nashville, Tennessee, USA, April 2006, pp. 467–474. 22. Wu XL, Lei S, Meng M, Cho JS, Lee SY. Coverage-driven Self-deployment for Cluster Based Mobile Sensor Networks. In Proceedings of IEEE International Conference on Computer and Information Technology Seoul, Korea, September 20–22, 2006 23. Wu XL, Yu N, Lei S, Cho JS, Lee YK, Lee SY. Relay shift based self-deployment for mobility limited sensor networks. In Proceedings of the 3rd International Conference on Ubiquitous Intelligence and Computing, Wuhan and Three Gorges, China, 2006. 24. Rappaport TS. Wireless Communications: Principles and Practice (2nd edn). Prentice-Hall: Upper Saddle River, HJ, 2002. 25. Xue WW, Luo Q, Chen L, Liu YH. Contour map matching for event detection in sensor networks. In Proceedings of the 2006 ACM SIGMOD International Conference on Management of Data (SIGMOD’06), Chicago, Illinois, USA, June 26–29, 2006, pp. 145–156. 26. Abadi DJ, Madden S, Linder W. REED: robust, efficient filtering and event detection in sensor networks. In Proceedings of the 31st international conference on Very large data bases, Trondheim, Norway, 2005, pp. 769–780. 27. Heinzelman W, Chandrakasan A, Balakrishnan H. Energyefficient communication protocols for wireless microsensor networks. In Proceedings of Hawaii International Conference on System Science, on the Island of Maui, January 2000. 28. Zhao L, Xu CN, Zhang TL, Xu YJ, Li XW. Cluster based energy efficient scalable resolution for wireless sensor network. The International Conference on Sensing, Computing and Automation (ICSCA 2006), ChongQing, China, May 8–11, 2006, Watam Press, pp. 2988–2933. Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
L. SHU ET AL. 29. Hsu CS, Tseng YC. Cluster-based semi-asynchronous powersaving protocols for multi-hop ad hoc networks. In Proceedings of International Conference on Communications (ICC2005), Volume 5, pp. 3166–3170. 30. Younis M, Youssef M, Arisha K. Energy-aware routing in cluster-based sensor networks. In Proceedings of the 10 IEEE/ ACM International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems, 2002, Fort Worth, Texas. 31. Hussain S, Matin AW. Hierarchical cluster-based routing in wireless sensor networks. In Proceedings of International Conference on Information Processing in Sensor Networks (IPSN 2006), 2006, Nashville, TN, USA. 32. Voigt T, Ritter H, Schiller J. Utilizing renewable energy in cluster-based sensor networks. In Proceedings of Swedish National Computer Networking Workshop, 2003, Stockholm, Sweden. 33. Kalpakis K, Dasgupta K, Namjoshi P. Maximum lifetime data gathering and aggregation in wireless sensor networks. In the Proceedings of IEEE International Conference on Networking (NETWORKS’02), August 2002, Atlanta, GA. 34. Lei S, Jie Y, Lee SY. ETRI: a dynamic packet scheduling algorithm for wireless sensor networks. In Proceedings of ETRSI 2004, Lisbon, Portugal, December 5–8, 2004. 35. Lei S, Ling WX, Jie Y, Lee SY. Maximizing system value among interested packets while satisfying time and energy constraints. In Proceedings of ICN 2005, April 17–21, 2005 Reunion Island. 36. Jie Y, Lei S, Ling WX, Cho JS, Lee SY. ETRI-QM: reward oriented query model for wireless sensor networks. In Proceedings of EUC 2005, Japan.
Lin Zhang received her B.M. degree in Electronic Business from Wuhan University of Technology, 2006 and is currently pursuing her M.S. degree in Digital Enterprise Research Institute, National University of Ireland, Galway.
Hui Xu received her B.S. and M.S. degrees in Electrical Instrumentation Engineering from Zhejiang University of China in 2001 and 2004, respectively. Since 2004, she has been pursuing Ph.D. degree in Ubiquitous Computing Lab (UCLab) of Computer Engineering, Kyung Hee University of Korea.
Xiaoling Wu received her B.S. and M.S. degrees in Electrical Engineering and Power Engineering from Harbin Institute of Technology (HIT), China, in 2001 and 2003, respectively. Currently, she is a Ph.D. student of Department of Computer Engineering at Kyung Hee University, Korea.
Authors’ Biographies Lei Shu received his B.S. degree in Computer Science from South Central University of Nationalities, China, 2002, and M.S. degree in Computer Engineering from Kyung Hee University, Korea, 2004, and currently is pursuing his Ph.D. in Digital Enterprise Research Institute, National University of Ireland, Galway.
Jie Yang received her B.S. in Automation from University of Science and Technology of China and M.S. in Computer Engineering from Kyung Hee University, South Korea, in 2004 and 2006, respectively. Currently, she is a Ph.D. student of Department of Computer Engineering at Kyung Hee University, South Korea.
Copyright # 2007 John Wiley & Sons, Ltd.
Manfred Hauswirth since July 2006 is the Vice-Director of the Digital Enterprise Research Institute (DERI), Galway, Ireland and Professor at the National University of Ireland, Galway (NUIG). He holds an M.S. (1994) and a Ph.D. (1999) in Computer Science from the Technical University of Vienna. From January 2002 to September 2006, he was a Senior Researcher at the Distributed Information Systems Laboratory of the Swiss Federal Institute of Technology in Lausanne (EPFL). Prior to his work at EPFL he was an Assistant Professor at the Distributed Systems Group at the TU Vienna. He is a member of IEEE and ACM.
Wirel. Commun. Mob. Comput. (2007) DOI: 10.1002/wcm
