QoS-Aware Relay Node Placement in a Segmented ... - IEEE Xplore

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2009 proceedings

QoS-aware Relay Node Placement in a Segmented Wireless Sensor Network Sookyoung Lee and Mohamed Younis Department of Computer Science and Electrical Engineering University of Maryland, Baltimore County (slee22, [email protected]) Abstract—In some applications of wireless sensor networks (WSNs) it may be necessary to federate a number of disjoint segments. Linking these segments may be subject to varying inter-segment quality of service (QoS) requirements. This paper presents an effective approach for federating these segments. The main idea is to place relay nodes (RNs) in order to establish intersegment connectivity with the least number of RNs while meeting the desired QoS requirements. Finding the optimal number and position of RNs is shown to be NP-hard and heuristics are thus pursued. The deployment area is modeled as a grid with equalsized cells. Each cell is evaluated based on the residual capabilities of RNs populated in the cell. The optimization problem is then mapped to finding the cell-based least cost paths that collectively meet the QoS requirements. The performance of the proposed approach is validated through simulation.

I. INTRODUCTION A growing list of applications has fueled research on WSNs in recent years [1]. Most notable are those in harsh environments, such as coastal and border protection, searchand-rescue and battlefield reconnaissance. By getting miniaturized sensors to operate unattended in such hostile setup, it would be possible to avoid the risk to human life. Since typically a sensor node is battery operated and has limited processing and communication capabilities, a large set of sensors are involved to ensure coverage and increase the fidelity of the collected data. Upon their deployment, the nodes are expected to form a network in order to share data and coordinate their action when participating in the execution of a task. However, WSNs in these setups are subject to damage that the network gets partitioned into disjoint segments. In addition, it is sometimes desirable to federate the service of multiple autonomous WSNs in order to achieve a mission or to react to unforeseen large scale events, e.g. earthquake, invasion, etc. In such scenarios, the segments are required to share data and orchestrate their action to perform a particular task. Moreover, the desired interaction among the various segments may be subject to distinct per-link QoS requirements such as wireless bandwidth, etc. When the partitioning is caused by damage in the network, the inter-segment QoS requirements may be just a byproduct of the damage due to the different volume of the generated traffic. When federating a set of autonomously-operating WSNs, each segment may have its own QoS requirement depending on the application served by the segment. In both cases, it is desirable to federate the individual segments by establishing inter-segment links that meet the QoS requirements. If sensor nodes are stationary, RNs can be deployed to fill the reachability gap. Given the cost and overhead involved in the deployment of the additional nodes, the number of relays should be minimized.

Placement of the minimal count of RNs is a NP-Hard problem [2] even without the consideration of QoS goals. To address such an optimization problem, this paper presents an algorithm for Optimized QoS-Aware Placement of relay nodes (OQAP). OQAP pursues centralized greedy heuristics and opts to reduce the number of relays required for establishing a connected inter-segment topology that meets the desired QoS goals. OQAP models the area of interest as a grid of equalsized squares (cells). The size of a cell is calculated based on a communication range of a RN such that any RN in one cell can communicate with other RNs located in neighboring cells. The problem is then mapped to identifying the cells that ought to be populated with relays so that the total number of deployed RNs is reduced. OQAP first introduces a cell-based profit function defined as the available relaying capacity in a cell. OQAP then employs a variant of the Dijkstra’s algorithm to find the least cost inter-segment path that meets the desired QoS. To our best knowledge, this is the first paper addressing QoS-aware placement problem of relays in a segmented WSN. This paper is organized as follows. Section II describes the considered system model. Related work is covered in Section III. The details and analysis of OQAP are provided in Section IV and V. Section VI concludes the paper. II. SYSTEM MODEL In this paper two types of segmented WSNs are considered. The first is a WSN that got partitioned due to major damage, e.g. inflicted by explosives in battlefield, as shown in Figure 1. The second is a set of autonomous WSNs, each of which operates independently and collaborates to perform a task. Each segmented network is assumed to involve a number of micro sensors with constraint on their onboard energy supply and communication range. A sensor probes its surroundings and forwards the findings to a command center over a multi-hop path. Meanwhile a relay is a more capable node with significantly more energy reserve and longer communication range than sensors. Intuitively, relays are more expensive. Therefore the number of deployed relays is to be minimized. RNs Segment B Segment A are assumed to have the same Segment C communication range “R”. A distance Damaged Area between every pair Command center (CC) of segments may be longer than twice the value of Segment D Segment E : An active sensor the R, and thus multi-relay paths Figure 1: Illustration of a segmented WSN

978-1-4244-3435-0/09/$25.00 ©2009 IEEE

due to damage


may be needed. The inter-segment communication is also subject to heterogeneous QoS requirements when federating distinct WSNs, or simply due to the nature of the transmitted data. The QoS requirements make the federation problem very complex. To simplify the presentation, QoS requirements are assumed to be additive, e.g. bandwidth. III. RELATED WORK Many variants of the RN placement problem have been pursued as a means for shaping the network topology to meet some desired performance goals; each takes a different view depending on the relationship between the communication ranges of sensors and relays and the objectives of the optimization formulation [3]. Published work on RN placement can be grouped into two categories. The first considers unconstrained setups and opts to just establish connectivity between end points [4]. In the second category either additional performance objectives are targeted [5][6] or higher degree of connectivity is to be achieved [7][8][9]. In [4], Lloyd and Xue opt to deploy the minimum number of RNs such that a sensor is able to reach at least one RN, and the inter-RN network is strongly linked. The authors crafted polynomial time approximation algorithms based on forming a Steiner Minimum Tree with minimum number of Steiner Points, which is NP-Hard, and employing a Geometric Disk Cover algorithm. Unlike OQAP, heterogeneous QoS requirements are not considered. Meanwhile, Wang et al. [5] and Hou et al. [6] consider constraints on network lifetime besides the connectivity. Wang et al. opt to find a minimum number of RNs to meet connectivity between sensors and a base-station (BS) while the lifetime constraint is satisfied. They place first-level relays (FLR) that are directly connected to sensors and then populate the additional second-level relays (SLR) at a distance from FLRs so that FLRs stay operational for the longest time. SLRs also can be allocated only to the FLRs that cannot reach a BS. The number of SLRs is then reduced by removing redundancy. Hou et al. split sensors into groups led by an aggregation-andforwarding node (AFN). In order to prolong the AFN’s lifetime, they deploy additional RNs between the AFNs and a BS. Finally the optimization problem is to find the best positions of RNs with a given energy budget and number of RNs. They do not consider the minimum number of RNs. Achieving higher connectivity in two-tier networks is the goal of [7][8][9]. Tang et al. strive to place the minimum number of RNs such that each sensor is connected to at least two RNs and the inter-RN network is 2-connected [7]. They divide the area into cells and find the position for a RN to be connected to all sensors in a cell and also other RNs in neighboring cells. The work is further extended by Hao et al. in [8] by simply identifying positions that cover the maximum number of sensors, at which RNs are virtually placed. Through an analysis of the inter-RN connectivity, RNs with most coverage are switched to real to form a 2-connected graph. Meanwhile, Han et al. [9] consider a heterogeneous WSN of sensors with different transmission radius. They opt to deterministically place the least count of additional RNs to

establish k-vertex disjoint paths. However, none of these approaches considers QoS requirements. IV. QOS-AWARE RELAY NODE PLACEMENT OQAP opts to populate the least number of RNs such that the desired QoS goals and the inter-segments connectivity both are satisfied. OQAP is described in detail in this section. A. Major OQAP Steps Given the high complexity of the RN placement optimization considering the additional QoS requirements, OQAP pursues heuristics. The following describes the major steps: Model a segmented WSN as a grid: OQAP models the area of interest as a grid of equal-sized squares (cells). The side of a cell equals R / 2 . The rationale is that a RN centered at one cell should be able to reach relays located at the center of neighboring cells. Each cell is identified by its row and column on the grid. A path Pij between segments i and j located at ci and cj respectively, has two attributes denoted as DIS(Pij) and QoS(Pij); its distance, i.e., the number of cells between ci and cj, and its desired QoS value that this link should provide. A higher QoS implies more relaying capacity and may thus require multiple RNs deployed in the individual cells. Also a longer distance requires more RNs to support connectivity. DIS(Pij) can be simply computed as max(|row(ci)-row(cj)|-1,|column(ci)-column(cj)|-1). Model the grid as a directed graph GQ: Next, OQAP constructs a cell-based directed graph GQ=(VQ, EQ). VQ is a set of all cells in the area of interest. VSEG is a subset of VQ which includes the cells only where segments are located. A cell (vertex) is weighted based on the collective capacity of the RNs that the cell hosts. For instance, the weight W(cx) of a cell cx equals to zero if no RN exists, otherwise W(cx) reflects the uncommitted capacity of the RNs in cx denoted as CRN(cx). All vertices in GQ have initially zero weight since no RNs are populated yet. EQ is a set of edges which represent a bidirectional commutation link between every pair of neighboring cells. No edges exist between non-neighboring cells. Initially the weight is set to the distance between the centers of the two neighboring cells, i.e., R / 2 for a vertical or horizontal edge, or R for a diagonal edge. For the first inter-segment path selection, the distance based weights allow factoring in the energy consumed in communication. Thereafter, a weight of an edge in one direction is determined by considering the residual capacity of the RNs in the end cell of the edge. Figure 2 shows the initial GQ for the grid according to Figure 1. Each segment is assumed to be at the center of a cell. OQAP then converts the RN placement problem into finding the least cost path between specific cells in GQ. Update GQ based on the residual capacity of RNs and QoS: OQAP is a greedy algorithm which strives to meet the connectivity and QoS requirements with the least count of RNs, one path at a time. The weights of edges in EQ are used for path selection. After the first inter-segment path selection, OQAP reconstructs the weights of edges using the formula (1) based on the uncommitted resources in each cell. It is worth


B

R/ 2

02

03

04

05

A

07

10 R / 2 11

12

C

S3 14

15

16

17

20

21

22

23

24

25

26

CC

D

31

32

33

34

35

36

37

40

41

42

43

44

45

4746 E

47

R/ 2

R

01

R/ 2

Figure 2: An initial distance-weighted graph GQ for Figure 1; segments A to E and CC are mapped into c06, c00, c13, c30, c46, and c27.

noting that all incoming edges to a cell have the same weight since the weight of an edge depends only on the end vertex.. if W ( c v ) ≥ QoS ( Pij ) ⎧ 0, ⎪ ………….(1) ⎪ W (cu , cv ) = ⎨ ⎞ ⎛ ⎪ ⎜ 1 − W ( c v ) ⎟ , Otherwise ⎟ ⎪ ⎜⎝ QoS P ( ) ij ⎠ ⎩

To illustrate, let us consider the segments i and j, located at ci and cj respectively. After finding the least cost path for Pij, OQAP then identifies every vertex cx along the path, whose W(cx) is less than QoS(Pij) and populates new RNs to meet QoS(Pij) in cx. After that, OQAP reduces the weight of all vertices in the path, by QoS(Pij) to reflect the uncommitted capacity. Before considering the next path Pab, OQAP updates a weight W(cu, cv) of each edge (cu, cv) in EQ with respect to CRN(cv). If the CRN(cv) is zero, or no relays have been placed at cv, then W(cu, cv) is set to 1. Otherwise, its weight is between 0 and 1 according to the formula (1). The rationale of the formula (1) is that the weight of all edges towards a cell is inversely proportional to the relaying capacity that the cell has. Thus, an edge to a cell with more residual capacity will be favored when establishing the next inter-segment path. This promotes utilization of existing resources and ultimately minimizes the count of relays. Find QoS-aware and resource-efficient paths: OQAP uses inter-cell proximity for the first path and the available relaying capacity at the individual cells thereafter. The question that comes to mind is in what order the segment pairs should be considered. One of possible criterions is based on the stringency of the QoS requirements, measured as the ratio of the average QoS value on all paths (QoSAVG) to the full capacity of a single RN (CRN). If QoSAVG is much less than CRN, the minimum number of RNs for meeting the QoS and connectivity goals is going to be more influenced by DIS(Pij) due to the limitation of the communication range. Therefore, when staring with the longest path, there is a higher potential for reusing the residual capacity of RNs in GQ. For the same reason, OQAP prefers to process paths with higher QoS demands if the QoSAVG is significantly larger than CRN. If the ratio of QoSAVG to CRN is small, paths are randomly chosen. In order to identify the least cost inter-segment path, OQAP employs a modified version of Dijkstra’s algorithm,

referred to thereafter as OQAP_Connect. OQAP_Connect works like the original Dijkstra’s algorithm except it terminates as soon as a destination is reached for the first time ignoring all other non-relaxed (unsearched) vertices in GQ. Thus OQAP_Connect significantly reduces searching time which may be serious as the number of cells grows. It is worth noting that OQAP_Connect returns the same shortest path as the original Dijkstra’s algorithm does and thus maintains its optimality (Proof is omitted due to space constraints). B. Pseudo Code and Illustrative Example In this section, the pseudo code of OQAP is provided and its execution is illustrated through a detailed example. The following is the definition of additional notations: NRN : Total number of RNs returned by OQAP NRN(Pij) : Total number of RNs required to support Pij only NRN(cx) : Number of additional RNs required in cx

Figure 3 shows the pseudo code for OQAP. OQAP first sorts a set of paths, SP={Pij| ∀ ci, cj∈VSEG} in a decreasing order of DIS(Pij) or QoS(Pij) depending on the ratio of QoSAVG to CRN. (lines 1-3). The lines 5-22 explain the iterative greedy process to find the shortest inter-segment path returning the least number of RNs based on GQ. Along the least-cost path returned by OQAP_Connect, OQAP sums the number of new relay nodes according to the formula (2) (line 16). Then OQAP populates the new NRN(cx) of RNs in cx and updates CRN(cx) in line 17. For a next path Pij, GQ is updated using Update_GQ(i,j) function then the same process is repeated. OQAP_in_SegmentedWSN() 1. IF CRN >> AVERAGE{QoS(Pij) for ∀ ci,cj ∈ VSEG} THEN 2. Sort SP in the decreasing order of DIS(Pij) 3. ELSE Sort SP in the decreasing order of QoS(Pij) ENDIF 4. NRN Å 0 5. Pij Å Remove the first path in the sorted SP 6. GQ Å the initial distance-only-weighted graph 7. WHILE there exist any P ij not processed yet DO 8. prev[] Å OQAP_Connect(GQ,i) 9. DeployedPij Å {} 10. cx Å prev[cj] 11. NNnRN Å 0 12. WHILE cx ≠ ci DO 13. DeployedPij Å DeployedPij ∪ { cx } 14. NNnRN Å NNnRN + NRN(cx) 15. CRN(cx) Å {CRN(cx)+(CRN* NRN(cx))} – QoS(Pij) 16. cx Å prev[cx] 17. ENDWHILE 18. NRN Å NRN + NNnRN 19. Pij Å Remove a next path in the sorted SP 20. GQ Å Update_GQ(i,j) 21. ENDWHILE 22. return NRN Update_GQ(i, j) 23. FOR each edge (cu,cv) ∈ EQ DO 24. IF CRN(cv) ≤ 0 THEN W(cu,cv) = 1 ELSE 25. Update W(cu,cv) using QoS(Pij) based on formula (1) 26. ENDIF 27. ENDFOR Figure 3: Heuristic greedy algorithm of OQAP


N RN ( Pij ) =

∑N

(c ) ,

utilization shows in reducing RNs count. Having a high Number of segments may increase the connectivity and QoS requirements. The performance is assessed with the following metrics: z Obviously the number of RN, is the objective of the optimization and sure captures the effectiveness of OQAP. z The Convergence time of OQAP primarily depends on the number of cells searched in the OQAP_Connect procedure.

where

RN x c x ∈ Path returned by OQAP _ Connect ()

………….(2)

0, if W ( c v ) ≥ QoS ( Pij ) ⎧ ⎪ ⎪ N RN ( c x ) = ⎨ ⎪ ⎡ QoS ( Pij ) − W ( c x ) ⎤ , Otherwise ⎥ ⎪⎩ ⎢⎢ C RN ⎥

z

Figure 4 shows an illustrative example for OQAP. Assume SP={PD,CC, PB,CC, PC,D}, QoS(PD,CC)=600, QoS(PB,CC)= 350, QoS (PC,D)=50, CRN=500, and R=300. QoSAVG is assumed to be much smaller than CRN, and thus the paths are processed in the decreasing order of distances. Since PD,CC and PB,CC have the same distance, PD,CC is selected randomly. The first path PD,CC is processed based on Figure 4(a) in which a weight of each edge is about 212 or 300. One of the shortest paths returned by OQAP_Connect for PD,CC is {D, c31, c32, c33, c34, c35, c36, CC}. Since QoS(PD,CC) is 600, 12 new RNs are populated and each of CRN(cx) becomes 400 along the path. OQAP then reconstructs GQ based on QoS(PB,CC) and CRN(cx), where ∀ cx ∈ VQ. In Figure 4(b) a weight of dotted links becomes zero since each of CRN(cx), x=31 to 36 is larger than QoS(PB,CC)=350. A weight of all other links becomes one. Based on the updated GQ OQAP_Connect returns {B, c10, c20, c31, c32, c33, c34, c35, c36, CC} for PB,CC, which requires only two more relays because of reusing the residual capacity of RNs populated for PD,CC. In Figure 4(c) GQ is updated for PC,D. OQAP_Connect returns {C, c22, c31, D} and populates only one new RN in c22 since CRN(c31) equals QoS(PC,D). After all OQAP returns a total of 15 RNs for satisfying a given setup.

B. Baseline and Optimal Approaches We compare the performance of OQAP to the Always Initial GQ based algorithm (AIG), which, like OQAP, pursues greedy heuristics. AIG sets the shortest inter-segment paths using the original Dijkstra’s algorithm based on the distance. It does not factor in the available capacities at the cells. Therefore, its running time strictly depends on the number of cells in GQ. Cells along the picked path are populated with sufficient count of RNs. Given the same setup described in Figure 4, AIG would employ 20 RNs to meet all QoS requirements. Moreover, the performance of OQAP is compared to a brute-force based solution that identifies the optimal RN count. Starting from low count, the number of relays is incremented by one until it is feasible to achieve the connectivity and QoS goals. For each considered RN count, all possible combinations for placing the available RNs are tried. Clearly this is very time consuming and cannot be practically applied for a large grid. Nonetheless, running it for small setups would help in qualifying the performance of OQAP. C. Simulation Results We have simulated the multiple configurations with various values of CRN and R. The R varies from 25 to 300 with interval 25 and CRN is selected from {250, 500, 750, 1000}. QoS values are randomly picked in [0, 500]. The results of the individual experiments are averaged over 10 runs. All results are subjected to 90% confidence interval analysis. As seen in Figure 5, OQAP outperforms AIG for all R and CRN values. OQAP returns a smaller number of RNs when CRN and R increase. This is because a bigger CRN can increase the utilization of a RN and a larger R can cover a longer distance with the same number of RNs. The performance advantage of OQAP over AIG is distinct for small values of R=25. Figure 6 shows the number of RNs grows as the area of interest increases regardless of R or CRN. The area is assumed to be a square with the length of its side shown on the x-axis.

V. EVALUATION The effectiveness of OQAP approach is validated through simulation. This section describes the simulation environment, performance metrics, and experimental results. A. Experiment Setup and Performance Metrics In the experiments, five segments are randomly located in a 1200m×1200m area. The following are the parameters used to vary characteristics of the network: z Communication range of relays (R) determines the number of cells in GQ and thus has the most influence on the performance of OQAP. z Elevating the QoS or lowering CRN will sure boost the number of deployed RNs. z In large size of the area of interest, segments become distant from each other and the effectiveness of OQAP in resource 212

B 212

300

212

300

300

300 212

10

212

300

300

212 04

300

300 212

300

300

212 03

12

212

21

212 02

300

212

11

300

20

212

01

C

300 212

22

212

300

300

300

300

212 23

300

212 24

300 212

16

300

300 212

25

17

300 212

26

212

1

B

07 300

300

212

15

300

212

A 300

300 212

S3 14

300

212

05

01

02

300

212 40

300 212

D

300

300

300

300

212

300 212

31

41

300

212

300

300 212

32

33

300

300 42

212

212

300

300 43

212

(a)

300

300

212

34

300

300 44

300

300

212

212

35

300

300 45

300

300

212

300 212

36

300

300 4746 E

300

A

CC

1 10

1 1

11

C

12

S3 14

15

16

17

01

02

03

04

05

A

07

12

C

S3 14

15

16

17

22

23

24

25

26

CC

1

1 0

1 0

150 10

1

11

0

0

20

1

21 0

212

D

31

40

41

22

0

23

24

25

26

CC

0 1 0

1 1

400

37

1

B

07

212

1

1

150 20

0

0 0

0

21 0 1

400

400

400

400

400

32

33

34

35

36

37

D

31

42

43

44

45

47 46 E

47

40

41

1

50

1 0

0

0 1 50

50

50

50

32

33

34

35

36

37

42

43

44

45

47 46 E

47

50

212 47

212

05

1

0

212

04

03

1

1

(b)

(c)

Figure 4: An illustrative example of OQAP;(a) is the initial GQ (R=300) for PD,CC, (b) and (c) show an updated GQ for each of PB,CC and PC,D.


However, OQAP is more effective with large R and CRN because of the higher utilization of RNs. As also demonstrated in Figure 5, higher capacity or longer range yield fewer relays. Figure 7 compares the performance of OQAP to AIG while varying the number of segments and the desired QoS goals. In Figure 7(a), the performance advantage of OQAP becomes more distinct as the number of segments increases, mainly due to increased utilization of RNs. As seen in Figure 7(b), AIG yields the same node count until QoS reaches CRN, 750 in this experiment. This is because AIG does not take advantage of the existing relay capacity. Therefore, the AIG results are mainly affected by the inter-segment distances. Meanwhile, OQAP promotes resource utilization and employs much fewer relays, especially when QoS is much less than CRN. OQAP’s performance matches that of AIG when QoS reaches CRN since the capacity of populated RNs is full used. Figure 8(a) shows the execution time of OQAP and AIG, measured in terms of the number of cells searched while varying R, x-axis, and CRN, different bars. Since in AIG GQ does not reflect the residual relaying capacity, the number of cells scanned during Dijkstra’s algorithm remains the same as CRN varies. However, OQAP processes significantly fewer cells, especially for large CRN, since it strives to increase resource utilization. With respect to R, both search fewer cells as R grows since the total number of cells in GQ decreases for larger R. In Figure 8(b) OQAP reaches the optimal solution as CRN gets closer to QoS. the optimality of OQAP is maintained with about 87.5% gap even as CRN grows.

relay capacities to avoid the deployment of additional relays as much as possible. The simulation results have demonstrated the effectiveness of OQAP; especially when relays with high capacity and long communication range are employed. Future work includes extending OQAP by considering additional QoS requirements and more complex situation in which a segment spans more than one cells. Acknowledgement: This work is supported by the National Science Foundation, contract # 0000002270. REFERENCES [1]

I. F. Akyildiz, et al., “Wireless sensor networks: a survey”, Computer Networks, Vol. 38, pp. 393-422, 2002. Kariv, O., & Hakimi, S. L., “AN algorithmic approach to network location problems. I: The p-centers,” SIAM Journal of Applied Mathematics, 37(3), 513-538, 1979. M. Younis and K. Akkaya, “Strategies and Techniques for Node Placement in Wireless Sensor Networks: A Survey,” Journal of Ad-Hoc Networks, (to appear). E. L. Lloyd and G. Xue, “Relay Node Placement in Wireless Sensor Networks,” IEEE Transactions on Computers, 56(1), pp. 134-138, January 2007. Q. Wang, K. Xu, G. Takahara, H. Hassanein, “Locally Optimal Relay Node Placement in Heterogeneous Wireless Sensor Networks”, Proc. of the 48th Annual IEEE Global Telecomm. Conf. (Globecom’05), St. Louis, Missouri, November 2005. Y. T. Hou, Y. Shi, and H. D. Sherali, “On Energy Provisioning and Relay Node Placement for Wireless Sensor Networks,” IEEE Tran. on Wireless Comm., 4(5), pp. 2579-2590, Sept 2005. J. Tang, B. Hao, and A. Sen, “Relay Node Placement in Large Scale Wireless Sensor Networks”, Computer Comm., special issue on wireless sensor networks, Vol. 29. pp. 490–501, 2006. B. Hao, H. Tang, G. Xue, “Fault-tolerant relay node placement in wireless sensor networks: formulation and approximation,” Proc. of the Workshop on High Performance Switching and Routing (HPSR’04), Phoenix, Arizona, April 2004. X. Han, X. Cao, E. L. Lloyd and C.-C. Shen, “Fault-tolerant Relay Nodes Placement in Heterogeneous Wireless Sensor Networks,” Proc. of 26th IEEE/ACM Joint Conf. on Computers and Comm. (INFOCOM’07), Anchorage AK, May 2007.

[2] [3] [4] [5]

[6] [7] [8]

VI. CONCLUSION This paper has presented OQAP, which pursues greedy heuristics for establishing the inter-segment connectivity which meets QoS goals with the least RN count in a partitioned WSN. OQAP promotes the utilization of existing

OQAP (Crn=750)

AIG (Crn=750)

OQAP (Crn=1000)

AIG (Crn=1000)

350 300

OQAP (R=25)

AIG (R=25)

250 200 150 100

200 150 100

OQAP (R=250) AIG (R=250)

20

75

250

100 125 150 175 200 225 250 275 300

500

CRN

750

1000

OQAP (Crn=500) AIG (Crn=500) OQAP (Crn=750) AIG (Crn=750)

40

OQAP (Crn=1000) AIG (Crn=1000) 20

OQAP AIG

50

50 40 30 20 OQAP AIG

10 0

0 3

5

7

Number of segments

9

1000

1200

100

200

300

400

500

600

QoS (R=150, CRN=750)

(a) (b) Figure 7: Number of RNs depending on the number of segments (a) or QoS (b)

400

600

800

1000

1200

Size of the area of interest (R=150)

45000

Convergence time to solution (# of cells searched)

. Number of RNs

100

800

(a) (b) Figure 6: Effect of the number of cells on performance

60

150

600

Size of the area of interest (CRN=750)

(a) (b) Figure 5: Number of RNs under varying R and CRN. 300

0 400

700

800

OQAP (Crn=250) AIG (Crn=250) OQAP (Crn=500) AIG (Crn=500) OQAP (Crn=750) AIG (Crn=750) OQAP (Crn=1000) AIG (Crn=1000)

40000 35000 30000 25000 20000 15000 10000 5000 0 25

50

75

R

100

125

10

Number of RNs .

50

R

.

AIG (R=175)

0

0 25

Number of RNs

OQAP (R=175)

40

AIG (Crn=250) 60

50

0

200

OQAP (Crn=250)

AIG (R=100) 60

250

50

250

80 OQAP (R=100)

.

AIG (Crn=500)

AIG (R=100) AIG (R=200) AIG (R=300)

.

OQAP (Crn=500)

OQAP (R=100) OQAP (R=200) OQAP (R=300)

400

Number of RNs

AIG (Crn=250)

.

300

OQAP (Crn=250)

Num ber of RNs

. 350

Num ber of RNs

80

450

400

Number of RNs

450

[9]

9 8 7 6 5 4 500

AIG OQAP OPTIMAL 700

900

1000

CRN (R=300)

(a) (b) Figure 8: Convergence time to solution for the various R (a) and comparison of OQAP and AIG to the optimal (b)