Towards efficient dynamic surface gateway deployment for underwater network

Ad Hoc Networks 11 (2013) 2301–2312

Contents lists available at SciVerse ScienceDirect

Ad Hoc Networks journal homepage: www.elsevier.com/locate/adhoc

Towards efficient dynamic surface gateway deployment for underwater network Saleh Ibrahim, Jun Liu ⇑, Manal Al-Bzoor, Jun-Hong Cui, Reda Ammar CSE Dept., University of Connecticut, Storrs, CT 06269, USA

a r t i c l e

i n f o

Article history: Received 6 November 2012 Received in revised form 26 March 2013 Accepted 28 May 2013 Available online 6 June 2013 Keywords: UWSN Gateway deployment Optimization framework

a b s t r a c t In underwater acoustic sensor network, deploying multiple surface-level radio capable gateways is an efficient way to alleviate the burdens of high propagation delay and high error probability during transmission. However, the locations of gateways need to be carefully selected to maximize the benefit in a cost-effective way. In this paper, we present our formulation of the surface gateway deployment problem as an integer linear programming (ILP) and we solve the problem with heuristic approaches to provide a realtime solution for large scale deployment problems. By applying the proposed heuristic algorithms to a variety of deployment scenarios, we show that they are nearly optimal for practical cases, which opens the door for dynamic deployment. Therefore, we extend our solution to a dynamic case and propose a modified framework that integrates Aqua-sim, a NS2-based underwater wireless sensor network simulator. Our simulation result shows the benefits of dynamic gateway redeployment over static deployment. Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction Recently, Underwater Wireless Sensor Networks (UWSN) have emerged as a new alternative technology enabling underwater monitoring and exploration applications, including scientific, commercial and military applications [1–5]. At the same time, it also brings many technical challenges due to its special character of acoustic communication, and water environment, such as long and propagation delay, high error probability, limited power supply [3–7]. One way to alleviate the effect of the high propagation delay introduced by acoustic communications in underwater sensor networks is to deploy multiple surface-level gateways. Fig. 1 illustrates an underwater sensor network with multiple surface gateways. In the sensor network, each sensor node can monitor and detect environmental

⇑ Corresponding author. E-mail addresses: [email protected] (S. Ibrahim), jul08003@ engr.uconn.edu (J. Liu), [email protected] (M. Al-Bzoor), [email protected] (J.-H. Cui), [email protected] (R. Ammar). 1570-8705/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.adhoc.2013.05.011

events locally and then transfer these measurements, through the network, to a surface gateway node (which is also referred to as a sink for the UWSN), which then relays data to the control station. Instead of having to use long underwater paths to reach a unique surface sink, a multiple-sink underwater sensor network differs from single sink networks in that nodes can send data packets towards a nearby surface-level gateway, as illustrated in Fig. 1. A surface gateway then uses radio waves to forward packets to the control station. Considering that electromagnetic wave propagation is orders of magnitude faster than acoustic wave propagation, it is safe to consider the radio propagation delay from a surface gateway to the control station negligible. The same can be said about energy consumption since acoustic communications consume much more energy than radio communications [5]. Accordingly, all the surface-level gateways (or sinks) along with the control station form one virtual sink. In our previous work [8], we studied in detail the factors and benefits of such an architecture and formulated the gateway optimal deployment as an integer linear programming (ILP) problem. In this paper, we demonstrate the

2302

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

Fig. 1. UWSN with multiple surface gateways.

performance of heuristic solutions to the gateway optimal deployment problem and show that high-performance solutions can be efficiently found in polynomial time, enabling the realtime solution of large-scale deployment problems and opening the door to investigating the dynamic re-deployment of gateway nodes. Therefore, we show how our gateway deployment optimization framework can be practically extended to a dynamic redeployment scenario by incorporating simulation-obtained performance parameters, and shows its benefit over static deployment. The work presented in this work is organized as follows: first, Section 2 presents related works and Section 3 introduces our general framework, then in Section 4, we show the heuristic approaches. Next, Section 5 demonstrates the dynamic deployment. Finally, Section 6 summarizes our contributions and suggests some directions for future work. 2. Related work In the literature, a variety of node deployment strategies for underwater sensor networks have been presented. [9,10] have identified two types of UWSN deployments. In a Two-Dimensional UWSN, sensor nodes are anchored to the bottom of the ocean. Underwater sensors may be organized in a cluster-based architecture, and be interconnected to one or more underwater gateways by means of a mainly horizontal acoustic data links. Underwater gateways relay data collected by sensors to a surface station through a mainly vertical acoustic link. In a Three-Dimensional UWSN, sensors float at different depths in order to observe phenomena that cannot be adequately observed by means of ocean bottom sensor nodes. These floating sensors can either be anchored to the sea bottom or hanging from surface buoys by means of variable length strings. The depth of a sensor node can then be regulated by releasing or retracting its anchor string. According to [9], optimal deployment of 2-D UWSNs involves determining the minimum number of sensors and underwater gateways to achieve a target sensing coverage and communication connectivity requirements. They also address the question of determining the surface deployment area in order that the sensor nodes land within a certain target bottom area. For 3-D UWSNs, the authors suggested three deployment strategies: 3D-random, bottomrandom, and bottom-grid. In a 3D-random deployment,

underwater nodes anchor themselves at random ocean bottom locations and then float at random depths. In bottom-random deployment, underwater nodes anchor themselves at random ocean bottom locations then information about their locations are used to calculate the optimal depth at which each node should float in order to satisfy sensing and communication coverage. In a bottom-grid deployment, a AUV is used to anchor underwater sensor nodes at predetermined locations, and each sensor is assigned its desired depth in order to achieve the target coverage ratio. The question of relay node deployment optimization has been the focus of a number of research efforts. [11] presented a MIP framework for describing the relay deployment optimization problem. Static link-scheduling and routing were used to control medium access to avoid collisions and to route data from sensors to the surface sink. In their formulation, the authors assumed that packets would have to travel from their respective sources to the sink within a single period of the schedule, resulting in unnecessarily long schedule periods. An interesting idea for improving UWSN energy efficiency is the use of mobile data collectors. The idea of mobile sinks or mobile data collectors has been proposed in literature both for underwater sensor networks [12,13]. When the application of the sensor network is not a realtime application, data can be stored at sensor nodes until a mobile data collector is in the vicinity. Data collectors make rounds traversing the sensor network following a path determined through an optimization process. A WSN architecture that relies on mobile data collectors have one main advantage, namely reducing the communication-related energy consumption of sensor nodes, thus prolonging the sensors lifetime. The redeployment issue has been addressed by some earlier research. [14] illustrated the need for redeployment due to the continuing drift of sensor nodes away from the required sensing coverage area due to their passive mobility. The authors assumed a random walk model and considered only replacing lost sensors as a corrective redeployment strategy. [15] addressed the problem of repositioning underwater sensors in order to improve the collective coverage of the network. In addition to minimizing the total cost of moving all sensors during the repositioning process, they also considered the problem of minimizing the maximal cost of repositioning any one set sensors assigned to a single AUV, in order to avoid depleting the feul of one AUV much earlier than others during the repositioning process. Our work differs from all the above in addressing the multiple sink deployment problem. It formulates the problem by taking delay and energy as the major objective functions with multiple cost and performance related constraints, and solves the problem using heuristic approaches and extends our solution to a dynamic case.

3. Problem formulation We apply the same problem setting described in [8], which assumes that there is a pre-existing underwater

2303

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

deployment and seeks to find the optimal deployment locations for a given number of surface gateways. The surface gateway deployment problem is formulated as an optimization problem. The formulation consists of a basic set of constraints that can be augmented with a variety of objective functions.

 All acoustic transceivers (including those which are part of surface gateways) are homogeneous and therefore the communication range is assumed to be constant for all nodes. We further assume that acoustic transceivers are using only one transmission power level.  The shared acoustic channel bandwidth is the same all over the network. Therefore, the capacity of all graph edges is the same and is equal to the channel maximum packet rate assuming a single sender.  Packets are of fixed length, and all transmissions are synchronized in fixed time slots. But reception times are not necessarily synchronized.  We assume that medium access is controlled by a link-scheduling mechanism.

3.1. Network model The surface gateway deployment problem is modeled as a graph optimization problem. The nodes of the graph represent underwater sensors and candidate surface gateway positions, and the problem is to choose a subset of the candidate surface gateway satisfying a set of flow conservation constraints, interference constraints and either constraints on the number of surface gateways or the required network performance. The selection of the candidate positions is sophisticated enough to be considered a separate problem on its own, and therefore it is deferred to future research. For the purpose of this work, the set of candidate surface points is considered as given and has to satisfy connectivity constraints as a pre-condition, i.e. each underwater node has to have a path to at least one candidate surface position, taking into account the communication ranges of the involved nodes. Associated with each underwater sensor node is a packet generation rate. Surface gateway nodes have to collect all generated data packets. Further assumptions are detailed in the next section. 3.2. Assumptions  We assume that sensor nodes are either stationary or that their motion is correlated strongly enough to assume that their relative locations are fixed.  The scenario being considered is that of a monitoring network, where most of the traffic originates at sensor nodes and travels through the network to the common sink station. Therefore, the analysis is limited to the, possibly multi-path, route from each underwater sensor to the virtual sink.  Under the assumption of homogeneity and ignoring queuing delays, it can also be assumed that packets traveling in the reverse direction will follow the reverse paths, and therefore its magnitude can be added to the traffic generated by the respective sensor nodes.  To reduce the problem complexity, we assume a simplified interference model, i.e. a node can transmit only when it is not receiving anything from its neighbors. Queuing delays, such as those caused by the MAC protocol, are ignored. This assumption can be realistic if the network is very lightly loaded, and the probability of collision is too small to affect the performance. The effect of limited channel capacity was investigated in [8].  The graph (of underwater sensor nodes and surface gateway candidate nodes) is connected, i.e., we can find a route from any underwater sensor node to the sink through at most one surface gateway.  Surface gateway nodes have to relay all generated data packets.

3.3. Definitions The network is modeled as a graph, in which nodes represent the underwater sensors and surface gateways, and edges represent pair-wise communication links. 3.3.1. Nodes Let U be the set of all underwater sensor nodes, and T be the set of candidate surface node positions. Let V be the set of all nodes, i.e.

V ¼U [T Let Cv be the set of nodes within the communication range of an underwater node u, i.e.

Cu ¼

n

v : v 2 V; v – u; dðu; v Þ 6 RCu

o

; 8u 2 U

where d(u, v) denotes the Euclidean distance between the two nodes u and v, and Ru denotes the maximum acoustic communication range of the underwater node u. 3.3.2. Edges Let the set of edges E be the set of all possible communication links, i.e.

E ¼ feðu; v Þ : u 2 U; v 2 Cu g Let E Ou denote the set of outgoing links of an underwater node u, i.e.

E Ou ¼ feðu; v Þ : v 2 Cu g; 8u 2 U Since surface gateways do not transmit data packets on their underwater acoustic interface,

E Ot ¼ /; 8t 2 T and E Iv denote the set of ingoing links to a node v, i.e.,

E Iv ¼ feðu; v Þ : eðu; v Þ 2 Eg; 8v 2 V 3.3.3. Data generation and link flow rates Let s be the packet transmission time, called the timeslot in slot. Let gu be the average packet generation rate at node u 2 U, i.e. the expected number of generated packets during the packet transmission time s, and let G be the total data generation rate of the entire network, which should

2304

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

be equal to the average packet arrival rate at the virtual sink, i.e. the number of packets expected to arrive at the virtual sink during the packet transmission time s,

G¼

X gu

ð1Þ

3.4.3. Per-node flow conservation constraints Flow conservation implies that for underwater sensor nodes, the sum of the average flows leaving a node equals the sum of the flows entering that node plus the local data generation rate.

u2U

Let fe be the average flow per packet time in edge e measured in packets per packet time. Let fvI be the average total flow into node v per packet time, i.e.

fvI ¼

X

fe ; 8v 2 V

e2E Iv

Let fuO be the average total flow out of node u per packet time

fuO ¼

X

ftO ¼ 0; 8t 2 T

0

otherwise:

;

8t 2 T

ð2Þ

3.4. Constraints The constraints can be classified into deployment constraints, flow conservation constraints and interference constraints. 3.4.1. Deployment constraints gateway Data can only be received at locations where surface gateways are deployed. This constraint can be written as follows:

ftI 6 xt G;

8t 2 T

ð3Þ

3.4.2. Interference constraints The simple interference model adopted in this paper assumes that a node cannot send while it is receiving, which implies that the total data transfer rate sent and received at any node cannot exceed the maximum capacity B of the communication link. This implies for underwater sensor nodes that

fvO þ

X

fuO 6 B; 8v 2 V

ð4Þ

u2I v

Because surface nodes do not transmit data packets underwater, the interference constraint formula for surface gateways reduces to:

X u2I t

fuO 6 B; 8t 2 T

3.4.4. End-to-end flow conservation constraints Flow conservation also implies that the total data generation rate of all underwater nodes must equal the total data absorption rate by the virtual sink composed of all surface gateways, since each packet generated by any source, must eventually be received by a surface node (before being relayed to the sink).

ð7Þ

3.4.5. Number of surface gateways The following constraint can be used to limit the number of gateways deployed to at most N:

X xt 6 N

3.3.4. Gateway presence indicator Let xt be a binary variables that indicates whether a surface-gateway is to be deployed at candidate location t, i.e.

1 if a node deployed at t;

ð6Þ

t2T

For gateways, there is no underwater outgoing flow and therefore:

xt ¼

8u 2 U

X ftI ¼ G

f e ; 8u 2 U

e2E O u



fuO  fuI ¼ g u ;

ð5Þ

ð8Þ

t2T

3.4.6. Minimizing expected end-to-end delay The problem can be solved for a choice of optimization goals. We arbitrarily choose the average delay as an optimization goal to illustrate the performance of the heuristic algorithms proposed here. The objective here is to minimize the expected end-toend delay for all packets. The end-to-end delay for a packet is the sum of the per-hop delay over the entire path from the source that generates the packet to virtual sink (i.e., the gateway) that receives it. The per-hop delay consists of three components: queuing and channel access delays, transmission time and propagation delay. Let Ku,t be the set of all active paths from source underwater node u to surface gateway candidate location t. let ki = (e1, e2, . . . , em) be one of the paths in Ku,t, where e1, e2, . . . , em are links that constitute the path ki. The end-to-end delay sEki of path ki can be written as

sEki ¼

X

se ; 8u 2 U; t 2 T ; ki 2 K u;t

ð9Þ

e2ki

Let fki be the rate of data flow from u to t through path ki. It follows that average end-to-end delay between u and t is:

P E h i k 2K fki ski E sEu;t ¼ Pi u;t ki 2K u;t fki The flow in each link in terms of the path flows can be written as follows:

fe ¼

XX

X

fki

ð10Þ

u2U t2T ki 2K u;t :e2ki

Let de;ki be a set of decision variables that indicate whether link e is part of path k, i.e.

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

2305

Fig. 2. Operation of the greedy algorithm.

 de;ki ¼

1; e 2 ki

ð11Þ

0; otherwise

Using Eqs. (11) and (10) can be rewritten as follows:

fe ¼

XX X

de;ki fki

ð12Þ

u2U t2T ki 2K u;t

The average end-to-end delay between u and the virtual sink (all surface gateway candidate location) is:

E

P

 E

P

su ¼ Pt2T Pki 2K u;t

fki sEki

ki 2K u;t fki

t2T

P

t2T

¼

P

ki 2K u;t fki

gu

sEki

; 8u 2 U

P P Since g u ¼ t2T ki 2K u;t fki . Now, the overall expected end-to-end delay can then be written as:

P E

E½s  ¼

u2U

P

t2T

P

P

ki 2K u;t fki

sEki

u2U g u

Using Eqs. (10), (11), (9) and (1), the overall expected end-to-end delay can be rewritten as follows:

P E½sE  ¼ ¼

u2U

P

t2T

P

ki 2K u;t fki

P

e2ki

se

G

X 1 XX X ðfki de;ki se Þ G u2U t2T k 2K e2E i

u;t

1 X XX X ðse de;ki fki Þ ¼ G e2E u2U t2T k 2K u;t

i

1X ðfe  se Þ ¼ G e2E And the corresponding objective function will be

(

1X Minimize ðfe  se Þ G e2E

) ð13Þ

Now, the deployment problem can be solved by solving the optimization problem.

4. Heuristics approach As known, integer Programming is one of the 21 NPcomplete problems [16]. The MIP formulation of the gateway deployment problem can easily be shown to be NPhard. Therefore, finding the optimal solution may not be feasible for large problems. Even modest problems with <50 candidate positions can take hours to solve. A closely related combinatorial optimization problem is the Facility Location Problem (FLP), also referred to in literature as the Warehouse Location Problem (WLP). The heuristic algorithms that have been developed for the FLP, and especially its capacitated version, are good candidates for our gateway deployment optimization Problem (GDOP). 4.1. Background of FLP Informally, the Facility Location Problem (FLP) is concerned with finding the optimal locations for building facilities which supply a certain commodity or service, given the geographical distribution of the demand for this commodity or service and the cost incurred by transporting the commodity or service between each of the candidate facility locations and each of the demand centers. It can easily be seen that the GDOP can be reduced to a FLP, taking the gateways as the facilities, the relaying of data as the service and the data generated by the underwater sensors as the demand. The following variants of the FLP can be defined. The Uncapacitated Facility Location Problem assumes there are no upper bound on the demand that can be satisfied by a given facility, and hence every user is expected to obtain the commodity or service from the nearest (least expensive) facility. In a Capacitated Facility Location Problem, there is an upper bound on the demand that can be satisfied by a given facility, therefore it is necessary to find a user-facility assignment that minimize the total transportation cost while respecting the capacity bound of each facility. The latter variant can further be divided into two versions. With Unsplittable Demand, each user has to be assigned to a single facility which satisfies its entire demand.

2306

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

With Splittable Demand, the demand of one user can be divided across any number of facilities. If the minimum end-to-end transmission cost between a sensor node and a gateway is assumed to be independent on the load distribution in the UWSN, the GDOP can be reduced to a capacitated facility location problem with splittable demand. In the rest of this section both greedy and greedy-interchange algorithms for solving the deployment optimization problems are presented along with a complexity analysis of both algorithms. 4.2. The greedy algorithm In the greedy approach, the near-optimal deployment of m gateways is developed by adding (to an already existing near-optimal deployment of m  1 gateways) a gateway at the location that minimizes the objective function (we choose the average delay as an example). Algorithm 1 lists our implementation of the greedy heuristic algorithm and Fig. 2 illustrates the operation of the algorithm. Algorithm 1. Greedy heuristic algorithm Let the original MIP problem instance be P, and initially let P 0 ¼ P. For j = 1, 2, . . . , m do: 1. Modify the ‘‘Number of Gateways’’ constraint in P 0 such that the RHS of Eq. (8) is j 2. Solve P 0 , and let X0 be the solution (if one can be found) 3. For each t 2 T such that x0t ¼ 1 add a constraint xt = 1 to P 0 4. Repeat until j = m

Note that, the Greedy method still solves the optimization problem using the integer linear programming (ILP) to satisfy the objective and constraints, but it uses a greedy strategy. At the start point, the method assigns the number of the gateways as ‘‘1’’, and then calls the ILP routine to find the best candidates location in terms of satisfying the

objective and constraints. After finding the best candidate, it will add a new constraint which states that this candidate has been selected. Then it increases the number of gateways as ‘‘2’’, and calls the integer programming routine again to find the second position to satisfy the objective and constraints, including the new added constraint, which means one more gateway location will be selected out of the remaining candidate location. With this greedy strategy, it loops up to get the maximum of allowed gateways by adding one new gateway at each step from the set of candidate location to the previous solution. The algorithm quits unsuccessfully whenever P 0 at step (2) is infeasible. This means that the greedy algorithm will need modification in order to handle the special case when one single surface gateway is unable to serve the entire underwater sensor network. If the reason of infeasibility is bandwidth limitations, the data generation rates gv of all underwater nodes v, are scaled by j/m, to obtain a feasible solution. If again the problem in infeasible, then the underwater network has to be partitioned and each partition has to be solved separately, and then the greedy algorithm takes over from j = the number of partitions + 1. 4.3. The greedy-interchange algorithm The interchange algorithm takes a solution and tries to improve it iteratively. At each iteration, the algorithm searches for a pair of candidate deployment locations, one active and one inactive, whose interchange is most beneficial, i.e., leads to the maximum decrease of the objective function. Our implementation of the greedyinterchange is slightly different from the greedy-theninterchange standard approach. Instead we start from a greedy partial solution, and allow at most one of the already selected candidate locations to be exchanged for a better unselected location at the same time a new location is added to the solution in a greedy manner. Algorithm 2 lists the details of the implementation of the greedy-interchange heuristic algorithm and Fig. 3 illustrates the operation of the algorithm.

Fig. 3. Greedy-interchange algorithm.

2307

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

OðN  M  f ðN; KÞÞ Algorithm 2. Greedy-interchange heuristic algorithm Let the original MIP problem instance be P, and initially let P 0 ¼ P. For j = 1, 2, . . . , N do: 1. Modify the ‘‘Number of Gateways’’ constraint in P 0 such that the RHS of Eq. (8) is j 2. Solve P 0 , and let X0 be the solution (if one can be found) 3. Augment P 0 with the constraint:

X t2T

xt P j  1

The greedy-interchange, as a compromise solution between the optimization and the greedy approach, has to evaluate



M 1



 þ

M

  1

2

þ ... þ



 þ

M1

  2

1 1 2   N1 MNþ2 2

 þ

M2 2

  3 1

1

which means that the run-time complexity of the greedy interchange algorithm will be

OðM 2 N2  f ðN; KÞÞ

;X 0t ¼1

4.6. Performance evaluation 4. Repeat until j = N

4.4. Local search algorithms This class of algorithms have the following form: 1. Start from an initial feasible solution X 2. Repeat Generate the set of neighbors of X using some neighborhood operator, let the set of neighbors be called X. Search in X for the solution value that corresponds to the best value of the objective function, and move X to that new solution Repeat until a local optimal solution is found or a certain tolerance is achieved. To avoid getting stuck in small plateaus, a randomization process such as simulated annealing can be employed. 4.5. Complexity analysis The Mixed Integer Programming formulation of the deployment problem is an NP-hard problem. To find the optimal solution in the general case we have to evaluate   M possible deployment, where M ¼ jT j the size of the N candidate locations set and N, the maximum number of gateways to be deployed. In each deployment plan, we have to solve the optimal routing problem. Solving the optimal routing problem for a fixed underwater deployment and any given gateway deployment takes O(f(N, K)), where K is the number of underwater nodes. Therefore, the run time complexity of the brute-force search for optimal gateway deployment is

OðM N f ðN; KÞÞ O(f(N, K)) is the complexity of solving the optimal routing problem for the given underwater deployment problem. The greedy approach, however, has to evaluate       M1 MNþ1 M þ  deployments, which 1 1 1 means that the run time complexity of the greedy algorithm will be

To study the quality of the solutions generated using the greedy approach and the greedy-interchange approach, we solve several problems using four methods, namely exhaustive search, randomized, greedy and greedy interchange and then report the percentage of quality loss (increased delay) for each approach. We also record the run-time for solving each problem using each algorithm and compare the results. 4.6.1. Simulation settings To calculate the quality of the baseline randomized deployment we randomly generate 100 solutions for each problem and then calculate the average value of the objective function as well as its standard deviation. Throughout our experiments the 0.001-confidence interval was found to be 6 ± 6% of the value of the mean. Throughout the experiments, we fixed the packet length L = 400 bits, the underwater acoustic propagation velocity t = 1.5 km/s, and the transmission power is set to a constant of 1 W s per packet time. The communication range for the underwater modems for all nodes is fixed at RC = 150. We also fixed the area of deployment to a square area of 600 m  600 m horizontal extent, and fixed the candidate gateway deployment positions to a 5  5 square mesh of points spaced 150 m apart. The depth of all underwater sensors is arbitrarily set to 100 m, such that each of the underwater sensors, regardless of its horizontal location, is within the communication range of at least one surface gateway candidate position, thus satisfying the connectivity requirement. This guarantees that problems can be made feasible by setting a large enough limit on the number of surface gateway nodes, N. The following two underwater deployment patterns were used:  Uniform Underwater Deployment: The uniform underwater deployment was chosen because the uniformity of the solution simplifies the process of verifying the results. The chosen underwater deployment consists of a 7  7 planar mesh of sensor nodes. The distance between two adjacent nodes is 100 m, and therefore the nodes cover the entire 600  600 m area, as shown in Fig. 4.

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

The data generation rate at each underwater sensor is set to 1 packet per second. The acoustic channel effective bit-rate B is varied among 5 kbps, 10 kbps and 50 kbps. Accordingly, the packet transmission time s = 80 ms, 40 ms and 8 ms respectively and the data generation rate g = 0.08, 0.04 and 0.008 packets per packet time, and the energy consumption per packet transfer pS = 80, 40 and 8 W s respectively. For the purpose of our experiment, we fix the channel capacity so that the network load is very light. Namely, we set g = 0.01 packets per packet time. 4.6.2. Simulation results The results for the uniform underwater deployment are shown in Fig. 6, and the results for the random underwater deployments are shown in Figs. 7–9. Here, the excess delay stands for delay over the optimal solution, which actually means the surplus subtracting the delay with optimal solution from the actual delay. So in the figures, the curve of optimal delay is ‘‘X’’ axis.

35

Random Greedy Greedy-Interchange

30

Excess Delay %

 Random Underwater Deployment: Similar to the Uniform Underwater Deployment, except that the 49 underwater sensor nodes are distributed at random within the 600  600 m underwater area, as shown in Fig. 5.

25 20 15 10 5 0 0

5

10

15

20

25

Number of Gateway Nodes Fig. 6. Performance evaluation on propagation delay with uniform underwater deployment.

60

Random Greedy Interchange

50

Excess Delay %

2308

40 30 20 10

800m

Surface Candidate Positions

700

0

Underwater Sensors at 100m depth

0

5

600

10

15

20

25

Number of Gateway Nodes

500

Fig. 7. Performance evaluation on propagation delay with random underwater deployment 1.

400 300 200

70

0

60 0

100

200

300

400

500

600

700m

Fig. 4. Locations with uniform deployment.

Excess Delay %

100

Random Greedy Interchange

50 40 30 20

800m

Surface Candidate Positions

700

10

Underwater Sensors at 100m depth 600

0 0

5

10

15

20

25

500

Number of Gateway Nodes 400

Fig. 8. Performance evaluation on propagation delay with random underwater deployment 2.

300 200 100 0

0

100

200

300

400

500

600

Fig. 5. Locations with random deployment.

700m

In both cases, the greedy approaches produces results are very close to the optimal solution, with the greedyinterchange doing even better. To summary, we have empirically shown that the greedy approach can produce near-optimal solutions to the

2309

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312 50

40

Excess Delay %

N ¼ jSj

Random Greedy Interchange

45

Mobility Energy Cost. Let eM s;t be the energy consumed by gateway s 2 S to move from its current location to candidate location t 2 T . Redeployment Indicator. Let ys,t be a binary variable that indicates whether gateway s 2 S will be moved to candidate location t 2 T , i.e.

35 30 25 20



15

ys;t ¼

10

1; s will be moved to t 0;

otherwise

;

8s 2 S;

t2T

5 0 0

5

10

15

20

25

Number of Gateway Nodes Fig. 9. Performance evaluation on propagation delay with random underwater deployment 3.

5.1.2. Constraints At most one gateway can be deployed at any candidate location:

xt ¼

X ys;t 6 1; 8t 2 T

ð14Þ

s2S

A gateway can only be moved to exactly one location gateway deployment problem in underwater sensor networks. We have also shown that the quality of the solution can be further improved by employing the greedy-interchange technique at the cost of increasing algorithm complexity by factor of M, where M is the size of the set of candidate gateway deployment locations. 5. Dynamic deployment A static gateway deployment assumes that the sensing mission specifications are accurately known prior to the deployment. It also assumes that the models used for estimating the behavior of medium access control and routing protocol are accurate. Moreover, a static deployment optimization process does not account for possible future changes in the environment, the mission requirements, or the locations and status of underwater nodes. When the performance deviates far enough from the optimal, a redeployment should be considered. However, the cost of the redeployment process should be included within the optimization framework in order to justify the cost of redeployment by an equivalent or greater gain in the performance of the UWSN. In this section, we propose a dynamic gateway deployment strategy to cope with changes and fine-tune deployments to improve the performance of the UWSN. Namely, we address the following problem. Given an underwater sensor deployment and a pre-existing surface-level gateway deployment, find the optimal redeployment operation in order to maximize the network lifetime. The cost of the redeployment is measured by the energy consumed by the mobile surface gateway nodes to relocated themselves to their corresponding new redeployment locations. By relocating themselves, the nodes reduce the communication energy. 5.1. Mobile gateway redeployment problem 5.1.1. Definitions Gateway Nodes. Let S be the set of gateway nodes to be deployed (or redeployed). Therefore, the number of gateways N is

X ys;t ¼ 1; 8s 2 S

ð15Þ

t2T

5.1.3. Optimization objective We take the network lifetime objective in Section 3 as an instance. The decay rate of underwater (immovable) node, u 2 U, remains unchanged. In addition, we define a set of new variables to represent the decay rate of any redeployable surface gateway nodes, s, as follows:

xs ¼

X

" yðs; tÞ

t2T

#

ps ; 8s 2 S es  eMs;t

ð16Þ

And we assert the lifetime constraint for those gateways as follows:

xs 6 X; 8s 2 S 5.2. Performance evaluation In this section, we use Aqua-Sim, which is a NS2-based underwater wireless sensor network discrete event simulator, in order to simulate the operation of an underwater network after the initial optimal gateway deployment. We monitor the gap between the expected behavior and the real behavior of the network, and when appropriate a redeployment is scheduled. There are many benefits of using AquaSim. The static deployment optimization framework makes certain assumptions about network workload, routing, medium access control performance, node mobility, among other things. Therefore, the accuracy of performance estimates calculated through the static UWSN deployment optimization framework is limited by nature. Using Aqua-Sim, we can fine-tuned our deployment optimization solutions by incorporating into the optimization model simulation-obtained performance parameters that reflect the dynamics of physical, data-link and network layers. We can also simulate a dynamic deployment optimization framework, which enables redeployment to cope with dynamic changes to the UWSN topology, node failure, mobility and varying work load.

2310

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

assumed to stay in the same location until instructed by the gateway deployment module to move to a new deployment location. All underwater nodes generate data packets at a constant rate and send them to a gateway node (possibly over multi-hop). In our simulation we used the average end-to-end delay as the performance metric that should be minimized. The duration of the simulation was set to 16,000 s and a redeployment was triggered at fixed intervals. In order to gauge the effect of the frequency of redeployment, the redeployment interval was varied between 125, 500, 2000, 8000 and 32,000 s, the latter corresponding to no redeployment scenario.

Fig. 10. Effect of redeployment frequency on UWSN performance.

5.2.1. Simulation settings A set of underwater nodes at fixed depth of 100 m, initially deployed in a regular mesh. The underwater nodes are passively mobile following either a random-walk model or a constant drift model. A set of gateway nodes initially located randomly on the surface. Gateway nodes are

5.2.2. Simulation results The results of the simulation are shown in Fig. 10. It shows the advantage of dynamically adjusting the deployment locations of gateway nodes over static deployment. As expected, the end-to-end delay decreases as the redeployment frequency increases, but the effect is more evident when the underwater nodes follow the constant drift mobility model. For the random walk mobility model, nodes tend to remain within the vicinity of their original locations with high probability, which keeps the deployment near optimality. For the constant drift, the rate of deviation of the deployment from optimality will depend on the drift velocity. As a result, frequent redeployment leads to up to 25% improvement in expected end-to-end delay compared to the static deployment case.

Table 1 GDO framework symbolic notation. Symbol

Definition

U T V N

The set of underwater sensor nodes The set of candidate surface gateway positions The set of all nodes, i.e. V ¼ U [ T Number of surface gateways The communication range for node u

RCu RIu d(v, w) Cu E E Ou E Iv gv G T he,t fe fvI fvO xt

t s

sPv ;w ev pSe pRv pLv pv kv

xv X

The interference range for a transmitting node u The Eucledian distance between nodes

v and w

The set of nodes within u’s communication range, ¼ fv : v 2 V; v – u; dðv ; wÞ 6 RCu g The set of all edges e = (v, w), such that v 2 U; w 2 C v The set of edges going out of node v, ¼ fðu; v Þ : ðu; v Þ 2 Eg; 8u 2 U The set of edges entering node v, ¼ feðu; v Þ : ðu; v Þ 2 Eg;

8v 2 V

The packet generation rate per packet time v 2 U P The total packet generation rate per packet time ¼ v 2U g v The periodic link schedule length (number of slots) Binary transmission indicator for edge e during timeslot t, he,t = 1,node v transmits to w during time slot t, where e ¼ ðv ; wÞ; 8v 2 U; e 2 E The average flow per packet time in edge e in packets per packet time. P The average flow into node v per packet time ¼ e2E Iv f ðeÞ P Total flow out of node v per packet time ¼ e2E Ov f ðeÞ Binary gateway presence indicator, =1,a gateway should be deployed at location t Propagation velocity of sound waves in water Packet transmission time = timeslot Propagation delay from v to w, ¼ dðvt;wÞ Initial energy level at node v Transmission energy required for one data packet over link e Reception energy per packet. For surface gateways, this includes the energy required to relay a packet to the sink over radio Node

v listening/sleeping average energy consumption per packet time

Total power consumption of node v (energy per packet time) ¼ pLv þ Lifetime of node v (in packet times) Energy decay rate of node v, xv = 1/kv The decay rate of the entire network



pRv fvI þ pS fvO



S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

6. Conclusion and future work The placement of radio-capable surface gateways promise a dramatic improvement of underwater acoustic sensor networks characteristics. In addition to the obvious advantage of reducing end-to-end delay, it also reduces the overall energy consumption of the network by replacing some of the power-hungry underwater links with more energy-efficient radio links, and improves connectivity and reliability by eliminating the single point of failure. Those various benefits of surface-level gateways are highly dependent on the placement of the gateways. In this paper, we developed various heuristic algorithms for efficiently finding a near-optimal solution to the NP-hard deployment optimization problem, and compared their performance and complexity. And we presented a mathematical formulation for the gateway redeployment problem, and integrated our gateway deployment optimization framework into Aqua-Sim, a NS2-based UWSN simulator, and used simulations to assess the practicality of our model for both static and dynamic deployment optimization scenarios. In the future, we plan to explore the deployment scheme which can deal with mobility and queuing delay, and we would also evaluate other performance factor like error ratio and energy consumption. Besides that, we are still interested in finding out how the deployment scheme affects the design of network protocols like MAC, routing, by evaluating parameters like end-to-end delay, throughput, energy consumption, etc. Appendix A. GDO framework symbolic notation

[10]

[11]

[12]

[13]

[14]

[15] [16]

2311

– First ACM International Workshop on Underwater Networks, vol. 2006, 2006, pp. 48–55. D. Pompili, T. Melodia, I.F. Akyildiz, Three-dimensional and twodimensional deployment analysis for underwater acoustic sensor networks, Ad Hoc Networks 7 (2009) 778–790. L. Badia, M. Mastrogiovanni, C. Petrioli, S. Stefanakos, M. Zorzi, An optimization framework for joint sensor deployment, link scheduling and routing in underwater sensor networks, in: WUWNet 2006 – First ACM International Workshop on Underwater Networks, vol. 2006, 2006, pp. 56–63. W. Alsalih, H. Hassanein, S. Akl, Delay constrained placement of mobile data collectors in underwater acoustic sensor networks, Proceedings – Conference on Local Computer Networks, LCN (2008) 91–97. W. Alsalih, H. Hassanein, S. Akl, Placement of multiple mobile data collectors in underwater acoustic sensor networks, Wireless Communications and Mobile Computing 8 (2008) 1011–1022. L. Bin, R. Fengyuan, L. Chuang, Y. Yaqin, Z. Rongfei, W. Hao, The redeployment issue in underwater sensor networks (2008) 5117– 5122. Z. Kone, E.G. Rowe, T.A. Wettergren, Sensor repositioning to improve undersea sensor field coverage, OCEANS 2007, pp. 1–6. R.M. Karp, Reducibility among combinatorial problems, in: R.E. Miller, J.W. Thatcher (Eds.), Complexity of Computer Computations, 1972, pp. 85–103.

Saleh Ibrahim works as a lecturer at the Computer Engineering Department at the Faculty of Engineeing Cairo University, where he teaches several undergraduate and postgraduate courses. He also serves as a co-advisor for a number of graduate students. He received his Ph.D. degree from the Unversity of Connecticut back in 2010, where he worked as a member of the Underwater Sensor Networks Lab. His current research interests include randomized and approximation algorithms, wireless ad hoc networks networks and information security.

See Table 1. References [1] G. Xie, J. Gibson, A Networking Protocol for Underwater Acoustic Networks, CS Department, Naval Postgraduate School, 2000. [2] J.G. Proakis, E.M. Sozer, J.A. Rice, M. Stojanovic, Shallow water acoustic networks, IEEE Communications Magazine 39 (2001) 114– 119. [3] I.F. Akyildiz, D. Pompili, T. Melodia, Underwater acoustic sensor networks: research challenges, Ad Hoc Networks 3 (2005) 257–279. [4] J. Heidemann, W. Ye, J. Wills, A. Syed, Y. Li, Research challenges and applications for underwater sensor networking, in: 2006 IEEE Wireless Communications and Networking Conference, WCNC 2006, vol. 1, 2006, pp. 228–235. [5] J.H. Cui, J. Kong, M. Gerla, S. Zhou, The challenges of building scalable mobile underwater wireless sensor networks for aquatic applications, IEEE Network 20 (2006) 12–18. [6] I.F. Akyildiz D.Pompili T. Melodia, Challenges for efficient communication in underwater acoustic sensor networks, Newsletter- ACM SIGBED Review (Special issue on embedded sensor networks and wireless computing Homepage archive) 1, 2004, 3–8. [7] J. Kong, J.H. Cui, D. Wu, M. Gerla, Building underwater ad-hoc networks and sensor networks for large scale real-time aquatic applications, in: MILCOM 2005: Military Communications Conference 2005, vol. 2005, 2005. [8] S. Ibrahim, J.H. Cui, R.A. Ammar, Efficient surface gateway deployment for underwater sensor networks, in: IEEE Symposium on computers and Communications (ISCC 2008) 2008, pp. 1177–1182. [9] D. Pompili, T. Melodia, I.F. Akyildiz, Deployment analysis in underwater acoustic wireless sensor networks, in: WUWNet 2006

Jun Liu received B.Eng in Computer Science in 2002 from the Wuhan University, Wuhan, China. He is currently pursuing his Ph.D and working as a research assistant at the Underwater Sensor Network (UWSN) Lab, University of Connecticut. His major research focus on time synchronization, localization, deployment for underwater acoustic networks, and also interested in operating system, cross layer design.

Manal Al-Bzoor is a Ph.D. student of Computer Science and Engineering at University of Connecticut. She has received her B.S. degree from Jordan University of Science and Technology – Jordan and her MS degree in Computer Engineering from University of Michigan-Dearborn in 2004. She was awarded two scholarships from Yarmouk University for Masters Study and for current PhD study. She worked as an Instructor at Computer Engineering Department – Yarmouk University Jordan (2006–2009). Manal current research interests are focused on underwater sensor networks, and modeling/ simulation of wide range distributed systems.

2312

S. Ibrahim et al. / Ad Hoc Networks 11 (2013) 2301–2312

Jun-Hong Cui received her B.S. degree in Computer Science from Jilin University, China in 1995, her M.S. degree in Computer Engineering from Chinese Academy of Sciences in 1998, and her Ph.D. degree in Computer Science from UCLA in 2003. Currently, she is on the faculty of the Computer Science and Engineering Department at University of Connecticut. Her research interests cover the design, modelling, and performance evaluation of networks and distributed systems. Recently, her research mainly focuses on exploiting the spatial properties in the modeling of network topology, network mobility, and group membership, scalable and efficient communication support in overlay and peer-to-peer networks, algorithm and protocol design in underwater sensor networks. She is actively involved in the community as an organizer, a TPC member, and a reviewer for many conferences and journals. She is a guest editor for ACM MCCR (Mobile Computing and Communications Review) and Elsevier Ad Hoc Networks. She co-founded the first ACM International Workshop on UnderWater Networks (WUWNet’06), and she is now serving as the

WUWNet steering committee chair. Jun- Hong is a member of ACM, ACM SIGCOMM, ACM SIGMOBILE, IEEE, IEEE Computer Society, and IEEE Communications Society. Her email address is [email protected].

Reda Ammar received his M.S. and his Ph.D. degrees in Computer Science from University of Connecticut, USA in 1981 and 1983 respectively. Currently he is the head of Computer Science and Engineering department at University of Connecticut. His research interests include Software Performance Engineering, Real-time Systems, Parallel and Distributed Computing, Performance Modeling & Analysis and Computer Aided Performance Engineering. He has published over 290 articles in journals and conferences. He is a senior member of Institute for Electrical and Electronics Engineers (IEEE). He served as a president, a vice president and a member of the board of directors in International Society on Computers and their Applications (ISCA) and also the Editor-in-Chief of the ISCA journal.