Resource Aware Information Collection (RAIC) in Ad hoc Networks

Transcription

1 Resource Aware Information Collection () in Ad hoc Networks Justin Lipman and Paul Boustead and Joe Chicharo Telecommunications and Information Technology Research Institute University of Wollongong Wollongong Australia {justin,paul,joe John Judge Motorola Australian Research Centre Lord Street, Botany, NSW, Australia Abstract In an ad hoc network it may be necessary for a single node to collect information about the local or global state of the network and the devices contained within it. This information may be used in network management, resource and service discovery, and routing protocols. Given ad hoc network constraints, this process of information collection must be efficient, scalable and distributed. Unlike sensor networks, sensing in an ad hoc network is of secondary importance and therefore should introduce as little overhead as possible. This paper introduces Resource Aware Information Collection (). is a distributed, two phase (setup and capture), resource aware approach to sensing the state of an ad hoc network from a source node (sink). An optimised flooding mechanism is used to both disseminate sensor request messages and setup a backbone of suitable nodes to relay replies back to the sink. compared to and Direct Response consumes less energy and introduces less overhead (received and transmitted messages) when performing a query of the network. We show that in an energy and user constrained environment, it is possible when querying the network to extend its lifetime. This is achieved by accounting for node resources (during setup and capture), thereby distributing the load of relaying to nodes most suitable. This allows to perform significantly more queries while reducing the load of sensing. is able to collect information from over 9% of the network up to 6 queries compared to 8 queries for and queries for Direct Response. I. INTRODUCTION The advent of portable computers and wireless networking has lead to large growth in portable computing due to the inherent flexibility offered. Most wireless networks are built around an infrastructure, where all communications goes through base stations that act as gateways between the wireless and wired network. However, there may be situations in which it is impossible to construct such an infrastructure. An ad hoc network is a collection of wireless mobile nodes forming a temporary network lacking the centralised administration or standard support services regularly available on conventional networks. Nodes in an ad hoc network may act as routers, forwarding packets. Ad hoc networks may undergo frequent changes in their physical topology. Mobile nodes may move, thereby changing their network location and link status. New nodes may unexpectedly join the network or existing nodes may unexpectedly leave, move out of range or switch off. Portions of the network may experience partitioning or merging, which is non-deterministic. An ad hoc network may operate in isolation or connected to a fixed network (Internet) via a base station (gateway). Ad hoc networks are characterised by low bandwidth, high error rates, intermittent connectivity (partitioning), limited transmission range, device power constraints and limited processing capabilities. Ad hoc networks may be employed in many aspects of daily life (including home and office networks) as well as more specific tasks such as military applications, disaster recovery and conference events. In these networks there may be a need for all-to-one protocols which allow for information collection - sensing of the state of the network or the nodes that form the network. This sensing could be used for service discovery, autconfiguration, network management, topology discovery or data retrieval. There is a parallel between this type of sensing in an ad hoc network and that of sensor networks []. However, ad hoc networks and sensor networks differ in their application, construction, characteristics and constraints. The number of nodes in a sensor network may be significantly higher in number and have higher density, be more prone to failure, have frequent topology change and more severe device constraints such as limited processing, limited power and low bandwidth communication. Additionally sensor networks are built specifically for sensing. Whereas in an ad hoc network, sensing is of secondary importance. For these reasons, protocols suitable to ad hoc networks are not necessarily suitable to sensor networks and vice versa. In ad hoc network literature there has been a strong focus on information dissemination protocols for routing (one-to-one) and flooding (one-to-all), but little on all-to-one mechanisms or sensing in ad hoc networks. This paper introduces, Resource Aware Information Collection (), a distributed, two phase, resource aware approach to sensing the state of an ad hoc network from a sink. The first phase (setup) utilises a resource aware optimised flooding mechanism to efficiently create a backbone of relay nodes throughout the ad hoc network. The backbone is used in the second phase (capture) to relay replies from nodes in the ad hoc network back to the sink in an efficient and resource aware manner. As information is transfered along

2 the backbone, relay nodes where possible use a reverse path utility calculated during setup to determine the next relay node to forward the information back to the sink. The mechanisms employed attempt to reduce problems associated with both flooding and sensing in wireless ad hoc networks in terms of power consumption, the broadcast storm problem, implosion, resource blindness and overlap [][]. This paper is organised as follows. Section describes published mechanisms for information dissemination and information collection in ad hoc networks and sensor networks. Section describes the Resource Aware Information Collection. Section evaluates the performance of and compares it with existing mechanisms. Section concludes the paper. II. INFORMATION DISSEMINATION AND COLLECTION Although protocols designed for sensor networks may not be fully applicable to ad hoc networks. There is existing work in the area that may be applied to the design of ad hoc network sensing protocols. In sensor networks there are needs for both effective and efficient information dissemination and collection mechanisms. Information dissemination forms an integral part of information collection as sensor network protocols make use of dissemination for various reasons: for structural organisation of the sensor network during the setup phase, as a means for sensors to advertise availability of information and for sinks to inform sensors of interest in information. Sensor networks are based upon wireless broadcast technology, therefore it is important that this dissemination process be efficient and take account of the broadcast nature. Likewise it is important that the collection process be efficient. Collection is the process whereby information is directed from the sensors back to the sinks. Flooding is a simple mechanism that may be used for both information dissemination and information collection in both ad hoc networks and sensor networks. However flooding is a one-to-all mechanism and along with the broadcast nature of wireless communication technology it introduces significant problems [][]: Broadcast storm problem: Flooding is extremely costly and may result in redundant broadcasts, medium contention and packet collisions. Implosion: Implosion is a situation where a node will receive duplicate packets from a source due to multipath propagation. Resource blindness: Flooding does not take account of the characteristics, constraints and state of devices. Therefore nodes which are low in power or heavily loaded will receive packets when other nodes may be more suitable. Overlap: Overlap implies that nodes may share common information, which if disseminated may introduce unnecessary and additional overhead. There is significant literature [][][][][6][7] on optimised mechanisms for flooding in ad hoc networks. Utility Based Multipoint Relay (UMPR) flooding [8] is an optimised flooding protocol that forms the basis of the sensing mechanism described in this paper. It builds upon ideas from Multipoint Relay (MPR) [] flooding, with UMPR allowing for an optimised flood to be resource aware. In brief, the UMPR algorithm is (for more detail see [8]): ) Upon receiving a broadcast, determine all -hop and -hop neighbours that did not receive the previous broadcast. ) Calculate a forwarding utility U f for each -hop neighbour. Select the -hop neighbour with the highest utility, remove any -hop neighbours that will hear its broadcast and add the -hop neighbour to the MPR list, removing it from the list of -hop neighbours. ) Repeat from step until all -hop neighbours are covered. A. Sensor Networks In [9], Sensor Protocols for Information via Negotiation (SPIN) is described. It is a set of adaptive protocols that attempt to address the deficiencies of flooding by negotiation and resource adaption. The SPIN protocol builds upon a simple idea that the performance of a sensor network may be improved by using mechanisms that advertise data by sending a concise description instead of the data itself. An interested sink may then request that data. In [], is proposed. In this approach a sink node initiates a broadcast and attaches its interest - a task description describing the information of interest to the sink. Each sensor node then stores this interest in a cache along with a time-stamp and gradient fields. As the interest is propagated throughout the network, nodes setup reverse gradients to the sink in a distributed manner. A sensor will send new data of interest back to the sink via the gradient path. Data may be aggregated at intermediate nodes. When a sink begins to receive data of interest, it must repeatedly re-broadcast interests in order to refresh and reinforce th e gradients from the sources. Both SPIN and directed diffusion are referred to as datacentric routing [9]. In SPIN, sensors advertise available information and the sink must respond. In, the sink advertises its interest and the sensors reply. In [], Low Energy Adaptive Clustering Hierarchy (LEACH) is described. LEACH attempts, in a distributed manner, to minimise energy dissipation by randomly selecting sensor nodes as cluster heads so as to spread the high energy cost of communicating with a base station to all nodes in the network. LEACH is a two phase protocol - setup and steady phase. In the setup phase all nodes choose a random number and compare this to a threshold value. If less than a threshold the node is a clusterhead. The new clusterheads then advertise (broadcast) their status to the entire network. Sensory nodes attach themselves to a clusterhead based upon

3 signal strength and inform the clusterhead of this attachment. Sensor nodes are alloted time to send data to a clusterhead based upon a TDMA approach. In the steady phase sensory nodes may send data to clusterheads which may aggregate all data received from sensory nodes before sending this data to the base station. After a period of time the network enters the setup phase again. In [], Sequential Assignment Routing (SAR), a set of algorithms that perform organisation, management and mobility management are proposed. SAR generates multiple trees where the root of each tree is a one hop neighbour from the sink. Each tree grows outward from the sink. Nodes with low QoS and low energy reserves are avoided when forming the trees. Nodes may belong to more than one tree, which allows nodes to chose a tree to relay sensory information back to the sink depending on a tree s additive QoS metric and energy resources. In [], the Topology Discovery algorithm (TopDisc) for network management in wireless sensor networks is proposed. The algorithm finds a set of distinguished nodes that contain local neighbourhood information. The information from these nodes is then used to construct the approximate topology. The algorithm organises the network into a tree of clusters (distinguished nodes form the clusterheads). The algorithm allows for efficient data dissemination and collection through data aggregation. The main priority of sensor networks is for the flow of information from sensors back to the sink. However in an ad hoc network, sensing is of secondary importance, for this reason, when sensing in an ad hoc network, we need to ensure that both setup and collection of data are performed in such a way as to have little effect on the normal operation of the ad hoc network. The start up phase in a sensor network occurs from a state of no knowledge (self organising), however in an ad hoc network we assume there is knowledge of neighbouring nodes acquired from a routing protocol or through the exchange of beacon messages. We make use of this extra information to setup a backbone in a manner that reduces power consumption and the broadcast storm problem. Additionally, during the setup phase nodes are allocated in a parent-child tree structure that minimises overlap of broadcast replies from child to parent when information starts flowing back to the sink. The sensing we envision in an ad hoc network is from a network management or discovery perspective. Therefore, the lifetime of the backbone is intended to last for one query of the network and to have minimal effect upon the normal operation of the network. However, in future work we will consider mechanisms to extend the life of the backbone given node mobility and failure. III. RESOURCE AWARE INFORMATION COLLECTION We introduce Resource Aware Information Collection (), a distributed, resource aware, two phase (setup and capture) approach that allows for efficient information collection from multiple nodes in an ad hoc network to a single Fig Setup Phase node acting as the sink (all-to-one). Such information may be used in network management, resource and service discovery, and routing protocols. Obtaining the state of the network from a single node s perspective may be used in decision making to determine network conditions or some future actions. The state of an ad hoc network may include node characteristics (internal battery power, device utilisation, services, equipment type and user constraints such as benevolence) and link characteristics (utilisation, bandwidth and QoS). A. Setup Phase Figure shows s setup phase. A sink (node ) initiates a resource aware optimised flood to disseminate a sensory request () message throughout the ad hoc network. Black nodes are those nodes selected as relays whereas grey nodes are not relays. Directed dashed lines represent the optimised broadcast flow of messages. The length of the lines are representative of the distance between the nodes. Relays are selected based upon their forwarding utility for rebroadcasting the message. This is achieved by using mechanisms described in Utility Based Multipoint Relay Flooding [8]. The use of an optimised flooding mechanism reduces the broadcast storm problem experienced when flooding. The forwarding utility U f (equation ) used to select relays is resource aware. This is important as node s selected as relays must be capable of continuing a flood. Relays also form the sensory backbone used in s capture phase and therefore should be capable of supporting the flow of information back to the sink. Resource awareness in terms of a node s internal battery power is expressed in equation and any user-based constraints described as benevolence (B). Other utilities which account for link or node stability, device load or other user defined parameters may also be used. U f = BU p U n ()

4 Fig.. Capture Phase Urp = Up =. Up =. Urp = Up =. Fig.. Urp =. Urp =. Urp =. Up =. R= R= Up =.6 Reverse Path Utility Urp =.7 Urp =. be rebroadcast. U rp specifies the utility of a relay to forward a message back towards the sink. A node s U rp is a function of its own internal remaining battery power (equation ), total number of parent relays of a lower sink degree denoted as R and the sum of all parent relay s reverse path utilities. U rp (i) = U p R (U rp (j)) () 6 U n (i) = U p (i) = + e Pi+s () unallocated hop neighbours of node i total hop neighbours of node i The utility U p (equation ) specifies the remaining internal power utility of a device. A sigmoid function is used to determine the utility as it provides a good estimate of the required behaviour (low utility and slow change at low power; sharp change in utility at medium power; high utility and slow change at high power). P i is the remaining internal battery power of a device and is mapped onto the sigmoid. To shift the sigmoid function accordingly, s is defined as half the range of P i. The neighbour utility U n (equation ) for a node i is equal to the number of unallocated nodes in the -hop pool that are neighbours of node i divided by the total number of neighbours of node i. Therefore node i s utility will decrease as its shared neighbours are allocated to other relays. Each node maintains a list of parent nodes and their sinkdegree (the hop distance from the sink) attached to the message. As a message is propagated throughout the ad hoc network by relays, nodes may receive more than one message and therefore have more than one parent. In order to aid in the discovery of possible parent relays, nodes may delay replying to or forwarding messages. This delay allows for a relay when calculating the reverse path utility to determine any neighbouring relays. Additionally it allows for a child node to discover other more ideal parent relays. The formation of the parent-child structure combined with the sink-degree creates a sensory backbone with a reverse gradient path from all child nodes back to their parents and ultimately back to the sink. The final formation of such a backbone is shown in figure as a solid line between nodes (,), (,,7), (,,8,,) and (,,). During the setup phase, relays calculate a reverse path utility U rp (equation ) which is attached to the message to () The reverse path utility is important as it allows for relays during the capture phase to select the best path (if more than one exists) back to the sink. This is similar in aim to the mechanism used in SARS [], where nodes are elected as relays to form a path back to the sink based upon their additive QoS. Sensors may then select a path with the best QoS and energy reserves when sending data back to the sink., however allows nodes to perform this operation on a hop by hop basis as the message is routed back to the sink. The reverse path utility can be seen in figure where the path (nodes: 6,,,) from node 6 back to the sink (node ) is shown by a thicker directed line. B. Capture Phase Figure shows s capture phase. The capture phase occurs when nodes begin replying to received messages by sending a unicast sensory reply () messages to their parent relay. As the number of child nodes may be significantly more than parent nodes. Child nodes chose the parent node to which they are closest when unicasting a message. In the setup phase (figure ) node receives a message from node and node 9 receives a message from node. However in figure, node has selected node as its parent relay and node 9 has selected node as its parent relay. This process of child nodes selecting their closest parent is important when combined with transmission power control. It allows for a reduction in a node s power consumption due to packet transmission and isolates the broadcast effects on neighbouring nodes when a node unicasts its message to its parent relay. To reduce the number of packets flowing back to the sink, relays may in a similar fashion to concasting [], implement various mechanisms, based on merge and time semantics (directives from the sink attached to the specifying the type of information requested), to conserve bandwidth through aggregation of received unicast replies from child nodes. Parent relays wait a specific amount of time for a reply from child nodes before timing out. This timeout period, T o ut,

5 Percentage of Nodes Sensed.8.6. Percentage of Ad hoc Network Sensed. Number of Queries Minimum Energy Nodes 8 6 Minimum Energy Nodes vs Number of Network Queries Queries Fig.. Percentage of network detected per completed query Fig.. Minimum energy nodes per completed query may be based upon some metric. In this paper we assign a timeout period that is function of the relay s sink degree as shown in equation. However, this has the effect of limiting the depth as the further from the sink the smaller the timeout period. We intend in future work to look at other mechanisms for controlling this timeout period. T out = TimeoutConstant. () SinkDegree Various optimisations are possible in. Relays may be more selective about the replies they require from child nodes. If a relay has received recent beacon messages from child nodes that contains timely and relevant information, it may not need to rebroadcast a. This however requires that any relevant information that may be required be attached to beacon messages. When collecting information from an ad hoc network, reliability is an important issue. It is therefore important that in the presence of node mobility and node failure be able to recover and reliably collect information from all nodes. Future improvements to : (i) if a node fails to unicast a reply to its parent, it may then select another parent (of a lower sink degree) from its parent list and attempt to resend the unicast reply to that parent. (ii) reliability may be increased by parent relays prompting child relays for a reply when none has been received. (iii) to improve the reliability of the flow of messages back to the sink, relays that may not have a reachable parent relay may utilise child nodes, where possible, to forward a message to a child node s lower sink degree parent (if one exists). IV. RESULTS A simulation was developed to compare the performance of with both a Direct Response [] mechanism and [] by initiating a query of the ad hoc network. Direct Response was chosen as it provides a brute force approach to sensing and is applicable to ad hoc networks where a routing protocol may be used to direct replies back to the sink. was chosen as it may be implemented using current ad hoc network networking technology and like is an aggregated sensing mechanism. In Direct Response a sink initiates an Depleted Nodes Depleted Nodes vs Number of Network Queries Queries Fig. 6. Depleted nodes per completed query optimised flood using MPR to disseminate a message. A reverse path back to the sink via the preceding node is maintained at each receiving node. A node upon receiving a message unicasts a message back to the sink via the reverse path determined during the flood. The purpose of the query is to collect from the network, all node identifiers at the node acting as the sink. The simulation is executed multiple times with a different seed and the sink randomly selected. At the completion of a query the sink verifies the collected information (node identifiers) with the ideal obtainable information. The final results are averaged and 9% confidence intervals are generated. The simulation generates a stationary random topology of nodes. Nodes have a maximum transmission range of meters. Time is divided into epochs. An ideal MAC layer is assumed. There is no medium contention nor hidden-node scenario within the simulation as it is assumed that during an epoch all nodes can complete their respective transmission. The transmission medium is error free. A bidirectional link between two nodes is assumed upon reception of a beacon message. A first order radio model [] is assumed. In this model the first order radio dissipates E elec = nj/bit to run the circuitry of a transmitter or receiver and a further ǫ amp = pj/(bit m ) for the transmitter amplifier. Equation 6 is used to calculate the costs of transmitting a k-bit message a distance d. Equation 7 is used to calculate the costs of receiving a k-bit message. The radios have power control and consume the minimal required energy to reach the intended

6 recipients. The additional costs (RTS/CTS/ACK) associated with unicasting messages as opposed to broadcasting messages are accounted for. E Tx (k, d) = E elec k + ǫ amp k d (6) E Rx (k) = E elec k (7) Nodes in the network are assigned a random amount of remaining energy up to joules. Additionally, each node has its own user-based constraints. These constraints inhibit a node from rebroadcasting a message should the node s remaining energy drop below a fixed level - referred to as minimum energy. A minimum energy node is allowed to reply to a messages with its own message, but not forward messages received from other nodes. This is similar to a user with a laptop who wishes to use the network and support the services of the ad hoc network. However, a user may not wish the laptop s battery supply to be fully depleted while supporting network services. Clearly, a user may desire to specify what services the laptop is able to support given specific battery levels of the laptop. Figures, and 6 show the simulation results for, Direct Response and over queries of an ad hoc network. Figure shows the performance in terms of collected information for each of the mechanisms. is shown to provide significant performance improvement over multiple queries of an ad hoc network. after successive queries of the ad hoc network is able to collect information from 8% of the network, compared with % for. Direct Response depletes the nodes in the network and therefore has significantly worse performance. Additionally is able to collect information from over 9% of the network up to 6 queries. This is quite significant compared to 8 queries for and queries for Direct Response. Figures and 6 show the number of nodes that enter the minimum energy state and those that become depleted due to excessive use. In figure 6 at the completion of all queries, has 8% less depleted nodes than. In figure at the completion of all queries, the number of minimum energy nodes with is % less than with. In the constrained environment described, Direct Response performs poorly as it is a brute force approach and therefore is not aware of the resources available, nor is it efficient at collecting information from the ad hoc network. In figures and 6, the initial growth (up to queries) of the number of minimum energy nodes and depleted nodes is quite dramatic for Direct Response. The number of depleted nodes reaching and minimum energy nodes reaching. At this point, the nodes in the ad hoc network are not able to support Direct Response which then fails to collect information from the ad hoc network. This accounts for the decreasing gradient for Direct Response in the graphs. child nodes only reply to their closest parent node. This is important as it helps to localise the effects of unicast Joules Packets Transmitted Total Energy Consumed vs Node Nodes (constant concentration) 8 6 Fig. 7. Energy consumed per query Packets Transmitted vs Node Nodes (constant concentration) Fig. 8. Packets transmitted per query replies from child nodes. The use of blind flooding in directed diffusion may be beneficial in the setup phase in a sensor network as it is a brute force approach to disseminating a request for information and allows for the formation of multiple paths back to the sink. However, blind flooding is resource blind and cannot determine if a node has sufficient energy to relay information back to the sink. A recovery mechanism may be used to try alternative paths, but this introduces additional overhead. The use of an optimised flooding mechanism in results in fewer paths back to the sink. However selects the most useful nodes during setup to be relays. During the capture phase, relays only unicast a reply to a parent relay that offers the path back to the sink based upon the reverse path utility determined during setup. This greatly improves s performance (figure ) and ability to query a constrained ad hoc network. Figures 7, 8 and 9 show the simulation results for, Direct Response and for a single query of an ad hoc network. The concentration of the nodes is kept constant at 66.7 nodes per square kilometre. In a network consisting of nodes, to perform a single query requires 9% less energy, 8% less packet transmissions and 88% less packet receptions than Direct Response. Compared to, to perform a query consumes 8% less energy, transmits 9% less packets and receives % less packets. provides significant performance improvements over Direct Response and which make it suitable for use in an ad hoc network where routing (not

7 Packets Recieved Packets Recieved vs Node Nodes (constant concentration) Fig. 9. sensing) is the main priority. Packets received per query V. CONCLUSIONS In this paper we have described, a distributed, two phase (setup and capture), resource aware approach to information collection in ad hoc networks. A node selected as a sink initiates an optimized flood of a sensory request packet throughout the ad hoc network. This process of dissemination is resource aware, in that the relay nodes are selected based upon their ability to partake in both rebroadcasting the sensory request packet and also in forming a reliable tree-based sensory backbone during the capture phase. Upon receiving a sensory request packet, a node selected as relay then collects unicast replies from child nodes and attempts to aggregate the responses before unicasting a reply to its parent. Nodes are able to select their parent based upon their distance and also upon their reverse path utility back to the sink node. greatly reduces problems associated with information collection in wireless networks in terms of power consumption, the broadcast storm problem, implosion, resource blindeness and overlap. We show that in terms of energy consumption, received and transmitted packets, performs significantly better than the Direct Response approach. Additionally compared to an aggregated response approach such as, shows significant performance improvement. More importantly, we show that in an ad hoc network which is both energy constrained and has simple user constraints, is able to significantly extend the life time of the network while performing multiple queries. This is due to the resource awareness during the setup and capture phases of. [] A. Qayyum, L. Viennot, and A. Laouiti. Multipoint relaying: An efficient technique for flooding in mobile wireless networks. th Annual Hawaii International Conference on System Sciences,. [] Hyojun Lim and Chongkwon Kim. Multicast tree construction and flooding in wireless ad hoc networks. In Proceedings of the rd ACM international workshop on modeling, analysis and simulation of wireless and mobile systems, pages ACM Press,. [6] J.E. Wieselthier, G.D. Nhuyen, and A. Ephremides. On the construction of energy-efficient broadcast and mulitcast trees in wireless networks. IEEE INFOCOM,. [7] Justin Lipman, Paul Boustead, and John Judge. Neighbor aware adaptive power flooding in mobile ad hoc networks. In Special Issue on Wireless Networks in the International Journal of Foundations of Computer Science (to be published),. [8] Justin Lipman, Paul Boustead, and John Judge. Utility-based multipoint relay flooding in heterogeneous mobile ad hoc networks. In Proceedings of the Workshop on the Internet, Telecommunications and Signal Processing (WITSP ), pages 7, Wollongong, Australia, December. [9] W. Heinzelman, J. Kulik, and H. Balakrishnan. Adaptive protocols for information dissemination in wireless sensor networks. In Proceedings ACM MobiCom 99, pages 7 8, Seattle, WA, 999. [] Chalermek Intanagonwiwat, Ramesh Govindan, and Deborah Estrin. Directed diffusion: A scalable and robust communication paradigm for sensor networks. In Proceedings ACM MobiCom, pages 6 67, Boston, MA,. [] Wendi Rabiner Heinzelman, Anantha Chandrakasan, and Hari Balakrishnan. Energy-efficient communication protocol for wireless microsensor networks. In Proceedings of the Hawaii International Conference on System Sciences, pages, January. [] K. Sohrabi, J. Gao, V. Ailawadhi, and G.J. Pottie. Protocols for self-organization of a wireless sensor network. In IEEE Personal Communications, Vol.7, Iss.,, pages 6 7, October. [] Budhaditya Deb, Sudeept Bhatnagar, and Badri Nath. A topology discovery algorithm for sensor networks with applications to network management. In In IEEE CAS workshop, September. [] Kenneth L. Calvert, James Griffioen, Amit Sehgal, and Su Wen. Concast: Design and implementation of a new network service. In Proceedings of 999 International Conference on Network Protocols, Toronto, Ontario, 999. REFERENCES [] Ian F. Akyildiz, Weilian Su, Yogesh Sankarasubramaniam, and Erdal Cayirci. A survey on sensor networks. IEEE Communications Magazine, pages, August. [] Yu-Chee Tseng, Sze-Yao Ni, and En-Yu Shih. Adaptive approaches to relieving broadcast storms in a wireless multihop mobile ad hoc network. In International Conference on Distributed Systems, pages 8 88,. [] B. Williams and T. Camp. Comparison of broadcasting techniques for mobile ad hoc networks. In Proceedings of MOBIHOC, June 9-.