A Novel Mechanism for Restoring Actor Connected ... - IEEE Xplore

IEEE ICC 2015 - Ad-hoc and Sensor Networking Symposium

A Novel Mechanism for Restoring Actor Connected Coverage in Wireless Sensor and Actor Networks Noman Haider1, Muhammad Imran2, Mohamed Younis3, Naufal Saad1, Mohsen Guizani4 1

Dept. of Electrical and Electronics Engineering, Universiti Teknologi PETRONAS, Malaysia. [email protected] 2 College of Computer and Information Sciences, King Saud University, Saudi Arabia. [email protected] 3 Dept. of Computer Science and Elect. Eng., University of Maryland Baltimore County, USA. [email protected] 4 Qatar University, Doha, 2713, Qatar. [email protected]

Abstract- Provisioning network survivability is especially crucial in wireless sensor and actor network (WSAN) because nodes deployed in hostile environments are prone to frequent failures. Failure of an actor significantly impact actor connected coverage which is essential for effective network operation. Existing mobility-based recovery schemes are either geared towards restoring inter-actor connectivity or area coverage. None of them consider sustaining actor coverage (i.e., having sensors reachable to actors) while restoring inter-actor connectivity. This paper presents RACE, a novel mechanism to Restore Actor Connected Coverage with reduced recovery overhead. RACE distinguishes critical/non-critical actors based on 2-hop information to better assess the scope of the failure and optimize the recovery procedure. Neighbors of a failed actor employ a cooperative failure detection scheme and only perform a limited-scale network reconfiguration to adopt any bereaved sensors left unreachable (uncovered by an actor) due to failure of a non-critical actor. In case a critical actor fails, RACE substitutes it with a non-critical neighbor that has the least impact on coverage (i.e., number of sensors). If it is necessary to engage critical actors in the recovery, RACE is recursively applied by relocating actors until a noncritical node is picked. Simulation results confirm the performance advantage of RACE compared to the best contemporary schemes. Index terms – Sensor and actor networks, Network Survivability, Actor connected coverage, Controlled and coordinated movement.

I.

INTRODUCTION

Recent years have witnessed growing research interest in wireless sensor and actor networks (WSANs) because of their ability to operate in unattended setups. This attribute makes WSAN specifically suitable for safety-critical applications that require autonomous and intelligent interaction without human intervention. Example applications include critical infrastructure protection, fire detection and containment, landmine detection and deactivation. Actors in WSAN receive event notifications from sensors and collaborate with each other to respond by performing appropriate actions [1]. Maintaining actor connected coverage is required in most WSAN applications because actors have to be aware of emerging events and collaborate with each other to identify the most appropriate set of actors that will provide coordinated response. For example, for disaster recovery, e.g., an earthquake, sensors report presence of survivors to actors in the vicinity. The actors such as fire extinguishing robots, cranes and ambulances should be engaged as rapidly as possible to extinguish a fire, lift rubbles and rescue trapped survivors. This requires sensors to be within the reach of at least one actor (i.e.,

978-1-4673-6432-4/15/$31.00 ©2015 IEEE

covered by an actor). Unlike most previous works that consider actor area coverage [2-3], we consider a different definition of actor coverage which is based on number of sensors [4]. Moreover, sustaining inter-actor connectivity is vital to provide optimal orchestrated response and synchronize their operations. Therefore, maintaining actor connected coverage is crucial for the successful operation of WSAN. Nonetheless, the harsh environmental conditions, energy depletion, physical damage and hardware failures can cause frequent node failures. The repercussions of an actor failure are much more significant as compared to sensors in WSAN. Due to low cost, redundant deployment of sensors helps in tolerating some node failures. However, an actor failure may leave sensors in the vicinity unreachable and thus result in significant loss of actor coverage. In the absence of an actor, sensors will be unable to report an event and hence action cannot be taken. Moreover, failure of a critical actor that serves as a cut-vertex in the inter-actor topology will partition the actors into disjoint blocks and disrupt the on-going network operation. Hence, the autonomous, unattended and resource-constrained nature of WSAN requires provisioning agile and lightweight mechanism to sustain operation. The WSAN ability to continue its services despite the presence of some failures is referred to as survivability and is required in mission critical applications. Provisioning network survivability is extensively investigated in multi-hop wireless networks [5] and, to some extent, in sensor networks [6]. Most of the existing recovery schemes involving mobile nodes, such as WSAN, are either geared towards restoring inter-node connectivity or area coverage [6]. None of them consider sensor-actor reachability (referred as actor coverage) while restoring inter-actor connectivity. Basically, when actors relocate some of the sensors become bereaved by not being able to reach any actor to report their findings. This paper opts to fill this technical gap and devise a survivability mechanism to restore actor connected coverage. It presents RACE, a novel algorithm that employs controlled and coordinated relocation to Restore Actor Connected Coverage with minimum recovery overhead. To prevent overreacting to non-serious failures, RACE distinguishes critical/non-critical actors based on 2-hop information. The neighbors employ a cooperative failure detection procedure to confirm and assess the scope of failure. To tolerate the failure of a non-critical actor, limited topology reconfiguration is performed to adopt any bereaved sensors. To recover from the failure of a critical actor, node relocation is pursued. To prevent further network partitioning and to mitigate

6383


any loss in actor coverage, RACE prefers to pick a non-critical neighbor with least number of impacted sensors to substitute a failed critical actor. Before relocating, an actor makes an announcement so that its reachable sensors can join the neighboring, possibly over multi-hop paths. If deemed unavoidable, RACE employs a critical neighbor in a recursive manner until a non-critical node is encountered. The simulation results demonstrate the effectiveness of RACE compared to contemporary solutions. To the best of our knowledge, RACE is the first mechanism that restores inter-actor connectivity while ensuring sensor-actor reachability. The rest of the paper is organized as follows. We discuss the system model and problem statement in Section II. Section III compares RACE to related work in the literature. RACE is described in detail in Section IV. Section V presents the simulation results. Finally, the paper is concluded in Section VI. II.

SYSTEM MODEL AND PROBLEM DEFINITION

RACE is pertinent to a WSAN that consists of randomlyplaced large number of sensors and relatively much fewer actors. After deployment, nodes discover each other, organize themselves into clusters [4] and form a connected inter-actor network. Both sensors and actors maintain their association for sensor-actor communication. Each actor is responsible for providing an appropriate response to events reported by sensors within its range. RACE requires each actor to maintain updated information (e.g., actor ID, location, etc.) for all its direct and 2hop neighbors. Actors are assumed to have the same capabilities, and to be able to move on-demand to serve an event, and such relocation does not affect sensor-actor reachability, i.e., cluster membership does not change. The repercussions of an actor failure primarily depend on the number of affected sensors within its cluster and the position of the actor in the network. An actor failure may cause a sensor in its cluster to be unreachable to any other actor, either directly or over sensor-based multi-hop path. In such a case the sensor becomes orphaned and cannot serve the network. Moreover, the negative effect may be due to the length of the multi-hop path over which a sensor can reach another actor. Some WSAN applications are time sensitive and the sensor data should be delivered to an actor subject to latency constraints, which impose a cap on the length of the sensor-actor data route. Obviously, the fewer the number of sensors in the cluster of a failed actor, the less the impact of the actor loss. For example, failure of an actor A1 has less effect on actor coverage compared to Af as shown in Fig. 1. While the failure of a non-critical actor may make some sensors unreachable, it will not affect inter-actor connectivity which is more crucial. Meanwhile, the damage of a critical actor has more impact on actor connected coverage because it causes partitioning the network into disconnected segments in addition

to the bereaved sensors. For instance, consider an inter-actor topology depicted in Fig. 1. The failure of non-critical actors such as A9 and A6 do not affect actor-actor connectivity and the impact will be limited only to the degraded actor coverage. However, the loss of a critical actor like Af split the network into disjoint segments in addition to the degraded coverage. Therefore, RACE determines the criticality of the failed actor, in order to limit the recovery overhead. It is important to note that RACE considers one failure at a time and assumes that no other actor fails during recovery. III.

RELATED WORK

Although supporting network survivability has been extensively investigated in the context of wireless and sensor networks [5-6], relatively few studies have considered WSANs [7]. Published schemes can be classified as proactive and reactive. The former provisions resources to tolerate a faulty actor beforehand. Proactive approaches either place additional relay nodes or form a bi-connected topology [8] to avoid network partitioning in case of node failures. Meanwhile, reactive recovery schemes respond to a failure after it is detected. Due to the autonomous and unattended nature of WSAN, reactive strategies have been deemed more appropriate in order better support network dynamics. Coordinated node movement has been dominantly exploited as the means for reactively handling actor failures [6,9]. Controlled movement based recovery schemes are either geared for restoring area coverage or inter-node connectivity. Schemes, such as [10-11], have used controlled mobility of nodes to re-establish strong inter-node connectivity. Unlike these approaches RACE restores connectivity while factoring in sensor-actor reachability. On the other hand, few studies have considered both area coverage and connectivity. To make for the failed node, C3R [12] employs back and forth repositioning of the neighbors of a failed node to provide temporal coverage and intermittent connectivity. C3R imposes high recovery overhead and is suitable as a temporary solution unless additional nodes are deployed. Some schemes cared for actor connected coverage. For example, the authors of [13] exploited post-deployment actor movement to spread them for better coverage while sustaining the inter-actor links. However, RACE pursues on-demand relocation and deals with actor failures which are not considered in these efforts. RECRA [2] prefers to employ non-critical nodes to recover from a critical actor failure and is based on movement and transmission power control. RACE does not assume that the actor communication range can be selectively extended. In addition, RECRA is suitable in dense deployment which is not expected in a WSAN with limited actor population. To minimize operational overhead, some of the localized and distributed approaches [2] maintain 1-hop neighbor information and used [14] to determine critical/non-critical nodes. However, the inaccuracy of determining critical nodes leads to unnecessary recovery overhead. On the other hand, PADRA [3] assess node criticality (i.e., whether it is a dominator or dominate) by determining a connected dominating set of the whole network based on 2-hop. RACE employs LASCNN [15] to distinguish critical/non-critical nodes since it has been shown to yield higher accuracy when using 2-hop information. IV. RESTORING ACTOR CONNECTED COVERAGE

Figure 1: Impact of a node failure on actor connected coverage in a WSAN segment.

To assess the impact of an actor failure and minimize the recovery overhead, RACE employs a localized algorithm to

6384


classify critical and non-critical actors in a distributed manner. This section explains RACE in detail. A. Determing Node Criticality and Detecting Failure A cut-vertex in a graph is the node whose removal causes the graph to split into multiple connected components. In the context of the inter-actor topology, a cut-vertex will correspond to a critical actor, whose failure leads to partitioning the topology into disjoint segments. Determining whether a node is a cut-vertex can be done using depth-firth search, which requires the knowledge of the entire network state. Although it is very accurate, such an approach is not suitable for distributed reactive recovery schemes due to the excessive messaging overhead, especially in dynamic application setups. On the other hand, highly localized schemes (i.e., 1-hop based) are more appropriate for large-scale dynamic networks to support scalability, resource optimization and frequent variations in topology [14]. The accuracy of discovering cut-vertices by these schemes is low because nodes have very limited knowledge about the network. Basically, some nodes are deemed as critical while they are not really cut-vertices. To minimize false alarms, RACE classifies critical/noncritical nodes based on 2-hop neighbor information to balance between the operational overhead for criticality assessment and its accuracy. After network initialization, actors exchange status updates with neighbors. Each actor employs a localized algorithm, e.g., LASCNN [15], to determine whether it is critical or non-critical. In addition to sharing its criticality status, an actor will include in its state update message the number of sensors that cannot be part of another cluster. This information will be based on the assessment of the individual cluster members and is made during the clustering process. In other words, a sensor may favor joining its current cluster because it can reach the actor directly as opposed to reaching the actor of a neighboring cluster over multi-hop. A sensor will be abandoned if it does not have a k-hop neighboring actor where 0 < k < maxk and maxk is an application-determined parameter. RACE exploits such information to efficiently tolerate an actor failure. In other words, from coverage perspective RACE elevates the criticality of an actor if many of its sensors are unable to join neighboring nodes in case of failure. Detection of node failure is done through the exchange of heartbeats messages among neighbors. RACE capitalizes on the broadcast nature of status updates instead of requiring explicit heartbeat messages. Considering the unreliable nature of wireless links, RACE prevents an actor from being wrongfully perceived as faulty by tolerating intermittent misses of updates. Failure of an actor is determined by the neighbors based on missing a number of consecutive updates. The recovery procedure triggered by the neighbors primarily depends on whether the failed actor is critical/non-critical. B. Recovery from Non-critical Actor Failure Since failure of a non-critical actor does not affect interactor connectivity and only leaves some sensors uncovered, i.e., cannot reach an actor, the focus of the recovery process is to maximize the number of relinked sensors. An intuitive, yet impractical, approach is to employ a full-fledge network reorganization so that the bereaved sensors can join some actors. However, this approach may take long time and unnecessarily disrupt on-going missions. Therefore, RACE opts to employ a limited-scale network reconfiguration that engages the neighbor actors to move and cover the bereaved sensors.

RACE adopts an intra-cluster relocation of the neighbor actors to move as much as possible towards the sensors that need to be covered in order to shorten the multi-hop path length [16]. The premise is that some of the bereaved sensors will be directly reachable to the neighbors and will join their cluster. Moreover, sensor based multi-hop paths will be considered subject to a constraint on the length due to data latency. However, actors coordinate their relocation using the approach of [17] in order to maintain inter-actor connectivity. For example, Fig. 2 shows a recovery scenario for the topology of Fig. 1 where A12 fails and its neighbors (i.e., A10 and A11) relocate themselves slightly towards the bereaved sensors while maintaining existing connections. Some of these bereaved sensors directly join (shaded in green) the neighbor actors while the rest connect themselves through multi-hop paths (e.g., shaded blue at 2-hop, yellow at 3-hop and red at 4-hop). It is important to mention that some of the bereaved sensors are considered as orphan if their path length does not satisfy the

Figure 2: The neighbors (A10 and A11) of a non-critical failed actor (A12) pursue coordinated intra-cluster relocation to adopt the bereaved sensors.

application constraints. C. Tolerating the Failure of a Critical Actor Restoring inter-actor connectivity is the primary concern in case a critical actor fails. An intuitive solution is to replace the failed actor with the new one. However, such a solution is not practical for mission-critical time-sensitive applications because it requires long time. Moreover, harsh environmental conditions may prevent external intervention and thus make this solution infeasible. Another possibility is to identify spare actors (if available) and move them to the place of a failed node. However, such a provisioned approach may not be viable since redundant deployment of actors is costly and often is not pursued. Moreover, maneuvering an actor over long distances is not efficient due to energy constraints. Therefore, RACE exploits the mobility of neighboring actors. Upon failure detection, the neighbors of a critical actor independently determine the most appropriate substitute to displace Af. This is because moving some [2] or all neighbors [11] towards Af may grow the scope of the recovery to a network wide reconfiguration which is undesirable due to the increased overhead. The neighbors independently identify the actor that can strangle further partitioning (to prevent cascaded relocations) and have least impact on actor coverage (i.e., abandoned/orphaned sensors). Therefore, a non-critical neighboring actor with least actor coverage impact will be preferred to replace Af. To break a tie, RACE picks a noncritical substitute An with the highest neighbor actors (i.e., degree) because such a node is more likely to have overlapped coverage. A least distance neighbor to Af is picked to minimize the movement overhead in case of a tie. Note that since each actor is aware of the status of its 2-hop neighbors, An will be uniquely and independently determine by each neighbor of Af.

6385


Before maneuvering, An informs its neighbor actors so that they can cover its affiliated sensors in a similar manner discussed earlier. An then moves to the position of Af and informs the sensors in Af’s cluster. For example, Fig. 3 depicts that the neighbors of Af (in the topology shown in Fig. 1) detect the failure and pick a 2-hop non-critical actor with least coverage impact perform recovery, A6 in this case. A6 sends a movement notification to the neighbors before moving. On receiving A6’s message, its neighbors (i.e., A4, A5, A7, A8 and A9) pursue intra-cluster relocation and adopt abandoned sensors of A6 in a similar way discussed earlier. In case a non-critical node among the neighbors is not available, RACE employs a critical actor Ac in the recovery, hoping that the neighbors of Ac will handle its departure efficiently. Basically, Ac will replace Af and RACE is then applied by the neighbors of Ac, considering its relocation as a failure. In other words, RACE will be recursively applied. The premise is eventually a non-critical actor will be found, e.g., a leaf node in the topology, and the recovery process terminates. The selection criteria of Ac among all critical neighbors is close to that of non-critical ones; the actor that impacts the fewest sensors is favored with travel distance used to break ties. To avoid being wrongfully perceived as faulty, Ac sends movement notification to the neighbor actors so that an appropriate substitute As can continue the recovery procedure until a noncritical neighbor is reached. The procedure to replace the moved critical actor As with a non-critical neighbor will be similar to what is discussed earlier. Fig. 4(a) demonstrates that a critical neighbor A3 among others takes the lead based on the least impact on sensors and moves to recover from a cut-vertex A6 in Fig. 3. Before moving, it sends a movement notification to the neighbors and covers the affected sensors upon reaching to the position of A6. To complete the recovery, A2 will replace A3 and A1 will displace A2 as shown in Fig. 4(b). The neighbors will adopt the abandoned sensors of A1 as mentioned earlier. D. Pseudocode of RACE The pseudo code for the key steps is presented in Fig. 5.

Figure 3: Recovery from a critical actor failure (i.e., Af) in Fig. 1 through a non-critical neighbor A6 with least coverage impact.

SelfAssessment (AN, 2HopInfo, CMI) 1 IF (2HopNeigh (AN) stay connected without AN) 2 ANÆIs2HopCritical(True) 3 ELSE 4 ANÆIs2HopCritical(False) 5 SensorCount = CoverageImpact (AN, NeighA, CMI, maxk) FDR (AN, 1HopNeighA) 6 IF (AN detects failure of AF NeighA or receives movement notification from AF) 7 IF (AFÆIs2HopCritical( ) = = False) 8 CoordIntraClusterRelocation (AN, CMI) 9 RecoverSensors (AN, SF, maxk) 10 ELSE 11 FS = FindBestSubstitute (2HopInfo) 12 IF (AN = = FS) 13 MovementNotification (AN, NeighA) 14 Replace (AN, AF) 15 Adopt (AN, SF) 16 ENDIF 17 ENDIF 18 ENDIF FindBestSubstitute (2HopInfo) 19 IF (there is a non-critical actor in 2HopInfo) 20 Pick a non-critical neighbor with least coverage impact, highest degree and least distance to AF 21 ELSE 22 Pick a critical neighbor with least coverage impact and least distance to AF 23 ENDIF Figure 5: Pseudo code for RACE algorithm

Actors perform SelfAssessment based on 2HopInfo to determine whether they are critical or non-critical (lines 1-4). They use cluster membership information (CMI) to determine the coverage impact (line 5). They employ a failure detection and recovery (FDR) procedure to detect and recover from a neighbor failure (lines 6-18). A recovery process is triggered by a neighbor actor AN if it either determines failure of AF or receives a movement notification from AF. If the failed/moved actor is 2-hop non-critical, then its neighbors (e.g., AN) will pursue coordinated intra-cluster relocation towards the abandoned sensors and adopt them (lines 6-9). However, if AF is critical, then the neighbors individually determine the most appropriate substitute based on 2HopInfo to restore connected coverage. Before replacing AF, the substitute AN sends a movement notification message so that the neighbor actors can adopt its abandoned sensors. Upon reaching its destination, AN covers the sensors that are affected by AF’s failure (lines 10-18). RACE prefers to pick a non-critical neighbor with least coverage impact as a substitute for a critical failed actor (lines 19-20). If deemed unavoidable, a critical neighbor with least coverage impact is picked to displace to the position of AF (lines 22-23). V.

RESULTS AND ANALYSIS

The performance of RACE is validated through simulation experiments. This section explains the experiment setup, performance metrics and analysis on experimental results.

(a)

(b)

Figure 4: Recovery from a critical actor failure (i.e., A6 in Fig. 3) when a noncritical neighbor is not available. a) A critical neighbor A3 displaces A6 and adopts its sensors which results in successive relocations. b) A2 replacing A3 and A1 relocates to A2.

6386

A. Experiment setup and performance metrics The experiments involve deployed randomly sensors and actors in an event region of 1000m x 600m. We have created WSAN topologies that consist of varying number of sensors (500-1300) and actors (20-100). After deployment, nodes organize themselves into clusters using the approach of [4] and actors employ the algorithm [15] to assess the node criticality based on 2-hop information. The transmission


range of actors is assumed to be 100m unless stated otherwise. The following metrics were used to evaluate the performance of RACE: • The coverage preservation percentage (CPP): This metric gauges the effectiveness of RACE to preserve coverage in terms of recovering abandoned and orphaned sensors despite failure of an actor. It is the percentage of the sensor node population that can report to actors (i.e., actor coverage) after recovery relative to the pre-failure level. • The total movement overhead: This metric measure the total distance moved by all the actors that got engaged in recovery. The metric reflects the efficiency of RACE in terms of energy dissipation and topology changes.

• The number of recovery participants: This metric reflects

the scope of the recovery and indicates the level of interruption to the network operation which is crucial in mission-critical time-sensitive applications.

In the experiments, the following simulation parameters were used to vary the WSAN configuration: • Sensor density (SN): The actor coverage is affected by the number of sensor nodes in the network. • Actor density (AN): This parameter affects the coverage (i.e., number of supervised sensors) and the inter-node connectivity (i.e., critical/non-critical). Increasing the number of nodes boosts the actor connected coverage. • Actor communication range (Rc): The coverage and travel overhead is influenced by the communication range of actors. Based on the above metrics, we evaluate the performance of RACE through comparison with DARA [10] and RIM [11]. Like RACE, both the approaches exploit controlled and coordinated relocation of existing nodes to recover from an actor failure. However, they do not consider recovery of sensors (i.e., actor coverage) and their procedure is different. DARA simply replaces the failed node with the neighbor that has the least node degree, and requires recursive execution of the algorithm to tolerate connectivity loss due to movement of critical nodes. On the other hand, in RIM all neighbors are moved towards the failed node until they become connected. The algorithm is recursively applied to repair broken links due to movement. B. Results and analysis In the experiments, we generated random topologies with varying sensor and actor density and actor communication range. The number of sensors and actors has been set to {500, 700, 900, 1100, and 1300} and {20, 40, 60, 80 and 100}, respectively. The communication range Rc of actors is changed among 50, 75, 100 and 125. When changing the sensor count NS, the number of actors NA, the communication range of actors are fixed at 40, and 100m respectively. The number of sensors is set to 800 and communication range of actors is kept at100m

while changing NA. While varying Rc, the value of NS and NA are set to 800 and 60 respectively. The averaged results of 30 independent experiments with different topologies are reported. All results are subject to 90% confidence interval analysis and stays within 5% the sample mean. Coverage Preservation Percentage: Fig. 6 demonstrates the impact on actor coverage, measured in terms of CPP relative to the pre-failure level, while changing the sensor count, actor density and communication range. Overall, RACE almost preserves the actor coverage and consistently outperforms the baseline approaches. This is because RACE employs the neighbors of a failed non-critical actor to pursue coordinated intra-cluster relocation to adopt bereaved sensors either directly or through multi-hop. Moreover, RACE prefers to engage a non-critical neighbor with least coverage impact (i.e., the number of sensors that cannot be part of another cluster) to substitute a failed/moved critical actor which results in preserving coverage. On the other hand, RIM performs worse in terms of CPP. This is because RIM moves neighbors of the failed actor inward which results in shrinking the inter-actor topology and leads to unsupervised sensors at the periphery. Since, DARA only engage one neighbor in the recovery, therefore, it performs better than RIM. However, its coverage loss is higher than RACE because it does not consider coverage impact while moving an actor. Fig. 6 (a) shows that the performance of RACE is slightly improved with increased sensor density. This is mainly attributed to the fact that more bereaved sensors become reachable to neighbor actors either directly or through multi-hop paths. On the other hand, the performance of DARA is slightly decreased because it prefers to engage least degree actors in recovery that is often critical nodes. Fig. 6 (b) demonstrates that the performance of all approaches improves with the higher actor density. This is due to increased sensor-actor reachability which is better exploited by the RACE. The performance of CPP slightly degrades while increasing the communication range as shown in Fig. 6 (c). This is because more sensors are affected due to failure and thus makes it difficult for some sensors to join neighboring actors even over multi-hop paths. On contrary, the coverage loss is higher for DARA and RIM as both the approaches are mainly concerned with connectivity restoration and do not consider coverage. Total distance travelled: Fig. 7(a-b) illustrates the distance travelled by all actors during recovery. Both plots in Fig. 7 clearly indicate that RACE outperforms the baseline approaches despite performing intra-cluster relocation. This is mainly because RACE determines 2-hop critical/non-critical actors and does not cause movement overhead (except intra-cluster repositioning) in case of a non-critical node failure. Moreover, RACE prefers to move a non-critical node which does not cause further partitioning and hence strangles cascaded relocations. Furthermore, a high node degree indicates that a critical node is more likely to have non-critical neighbors; therefore, RACE limits the cascaded relocations by moving such a node. Fig. 7

(a) (b) (c) Figure 6: Coverage preservation percentage as a function of (a) number of sensors (b) number of actors (c) actor communication range.

6387


(a) (b) (c) (d) Figure 7: Total distance travelled by all recovery participants as a function of (a) actor density (b) communication range. Number of recovery participants as a function of number of actors in (c) and actor communication range in (d).

(a) indicates the scalability and consistency of RACE which is not much affected due to actor density. This is because RACE prefers to engage non-critical nodes in recovery. Similar observation can be made for the communication range (Fig. 7 (b)) where movement overhead is less than DARA and RIM. This is because RACE capitalizes the availability of non-critical actors due to availability of alternative links. On contrary, DARA and RIM does not prevent cascaded relocations and actors have to move over longer distances due to increased transmission range. From distance overheads point of view, RIM performs better than for small rc. Number of recovery participants: Fig. 7(c-d) reports the number of recovery participants (i.e., actors) for RACE and the baseline approaches. Both plots clearly indicate that RACE minimizes the number of moved nodes compared to DARA and RIM. This is mainly because RACE prevents overreaching to the failure of non-critical node. Moreover, RACE limits the scope of recovery and prevents cascaded relocations by moving a non-critical neighbor to substitute a failed/moved critical actor. While varying the actor density (Fig. 7 (c)) and communication range (Fig. 7 (d)), the performance of RACE scales well and remain consistent. On the other hand, RIM does not analyze the impact of failure and may engage all neighbors in recovery. Hence, widens the scope of recovery. But, RIM limits recovery overheads for small scale networks as compared to DARA.

REFERENCES [1] [2]

[3]

[4]

[5] [6]

[7] [8] [9] [10]

VI. CONCLUSION This paper has presented RACE; a novel survivability mechanism based on controlled and coordinated movement of neighbors to recover actor connected coverage with minimum overhead. Coverage in this context refers to the reachability of sensor to actors over a path that does not exceed a certain hop count. To prevent overreacting by responding to non-critical failures, RACE classifies critical/non-critical actors based on 2hop information for improved accuracy. The neighbors detect failure of non-critical actor and only pursue coordinated intracluster relocation to better adopt any sensors left unsupervised. To prevent further partitioning and minimize recovery overhead, a non-critical neighbor with least coverage impact is favored to replace the failed critical actor which also minimizes coverage loss and avoid large-scale network re-organization. Movement of a critical actor requires recursive execution of RACE and triggers cascaded relocations. The performance of RACE is validated through simulation. The simulation results have confirmed the performance advantage of RACE compared to contemporary schemes. ACKNOWLEDGMENT

[11] [12] [13] [14]

[15]

[16] [17]

This work was supported by the Research Center of College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia, through the Research Project no. RC121229.

6388

I. F. Akyildiz and I. H. Kasimoglu, "Wireless sensor and actor networks: research challenges," Journal Ad Hoc Networks, vol. 2 (4), pp. 351-367, 2004. M. Imran, M. Younis, N. Haider, and M. A. Alnuem, "Resource efficient connectivity restoration algorithm for mobile sensor/actor networks," EURASIP Journal on Wireless Communications and Networking, vol. Nov, 2012:347. K. Akkaya, F. Senel, A. Thimmapuram, and S. Uludag, "Distributed Recovery from Network Partitioning in Movable Sensor/Actor Networks via Controlled Mobility," IEEE Transactions on Computers, vol. 59 (2), pp. 258-271, 2010. K. Akkaya, F. Senel, and B. McLaughlan, "Clustering of wireless sensor and actor networks based on sensor distribution and connectivity," Journal of Parallel and Distributed Computing, vol. 69 (6), pp. 573-587, 2009. O. M. Al-Kofahi and A. E. Kamal, "Survivability strategies in multihop wireless networks," IEEE Wireless Communications, vol. 17 (5), pp. 7180, 2010. M. Younis, I. F. Senturk, K. Akkaya, S. Lee, and F. Senel, "Topology management techniques for tolerating node failures in wireless sensor networks: A survey," Computer Networks, vol. 58, pp. 254–283, January 2014. S. Sedighian Kashi and M. Sharifi, "Connectivity Weakness Impacts on Coordination in Wireless Sensor and Actor Networks," IEEE Communications Surveys & Tutorials, vol. 15 (1), pp. 145 - 166, 2013. S. Das, H. Liu, A. Nayak, and I. Stojmenović, "A localized algorithm for bi-connectivity of connected mobile robots," Telecommunication Systems, vol. 40 (3), pp. 129-140, 2009. M. Younis and K. Akkaya, "Strategies and techniques for node placement in wireless sensor networks: A survey," Journal Ad Hoc Networks, vol. 6 (4), pp. 621-655, 2008. A. A. Abbasi, M. Younis, and K. Akkaya, "Movement-Assisted Connectivity Restoration in Wireless Sensor and Actor Networks," IEEE Transactions on Parallel and Distributed Systems, vol. 20 (9), pp. 13661379, 2009. M. Younis, S. Lee, and A. A. Abbasi, "A Localized Algorithm for Restoring Inter-node Connectivity in Networks of Moveable Sensors," IEEE Transactions on Computers, vol. 59 (12), pp. 1669 - 1682 2010. N. Tamboli and M. Younis, "Coverage-aware connectivity restoration in mobile sensor networks," Journal of Network and Computer Applications, vol. 33 (4), pp. 363-374, 2010. K. Akkaya and S. Janapala, "Maximizing connected coverage via controlled actor relocation in wireless sensor and actor networks," Journal Computer Networks, vol. 52 (14), pp. 2779-2796, 2008. M. Jorgic, I. Stojmenovic, M. Hauspie, and D. Simplot-ryl, "Localized algorithms for detection of critical nodes and links for connectivity in ad hoc networks," in the Proceedings of the 3rd Annual IFIP Mediterranean Ad Hoc Networking Workshop (Med-Hoc-Net), Bodrum, Turkey, 2004, pp. 360-371. M. Imran, M. A. Alneum, M. S. Fayed, and A. Alamari, "Localized Algorithm for Segregation of Critical/Non-critical Nodes in Mobile Ad Hoc and Sensor Networks," in the proceedings of the 4th International Conference on Ambient Systems, Networks and Technologies (ANT2013), Nova Scotia, Canada, 2013. K. Akkaya, M. Younis, and M. Bangad, "Sink repositioning for enhanced performance in wireless sensor networks," Journal Computer Networks, vol. 49 (4), pp. 512-534, 2005. J. English, M. Wiacek, and M. Younis, "CORE: Coordinated Relocation of Sink Nodes in Wireless Sensor Networks," in the Proceedings of the 23rd IEEE Biennial Symposium on Communications, Kingston, Ontario, Canada, 2006, pp. 320-323.