Extending the Lifetime of Ad Hoc Wireless Networks

Transcription

1 Extending the Lifetime of Ad Hoc Wireless Networks R.I. da Silva 1, J.C.B. Leite 2, and M.P. Fernandez 3 1 Fac. de Tecnologia e Ciências de Conselheiro Lafaiete, Univ. Prof. Antônio Carlos Conselheiro Lafaiete, MG, Brazil roneilidio@bol.com.br 2 Instituto de Computação, Universidade Federal Fluminense Niterói, RJ, Brazil julius@ic.uff.br 3 Lab. de Redes de Comunicação e Segurança, Universidade Estadual do Ceará Fortaleza, CE, Brazil marcial@larces.uece.br Abstract. This paper presents a new algorithm, Extra, for extending the lifetime of ad hoc wireless networks. Extra tries to conserve energy by identifying and switching off nodes that are momentarily redundant for message routing in the network. Extra is independent of the underlying routing protocol and uses exclusively information that is collected locally. Simulation studies conducted have shown promising results. 1 Introduction Ad hoc wireless networks are computer networks without a predefined topology, whose nodes can move and communicate through radio channels, and cooperate to allow message routing. Several routing protocols for ad hoc networks were proposed in the literature, e.g., DSR [2] and AODV [5]. There are many applications for those networks, like support to rescue teams in disaster areas. An important characteristic of the mobile devices in an ad hoc network is the energy consumption, because they usually depend on batteries. In that way, besides the traditional metrics to evaluate a protocol, like packet delay and drop rate, lifetime of the network is important. The nodes in an ad hoc network consume energy not only when they are transmitting or receiving messages, but also when they are ready to listen for any data (idle state). That happens because the electronics of the radio has to be energized to maintain the capacity of receiving messages. Several studies indicate that the power for transmission, reception and listening is, typically, in the order of 1.40W, 1.00W and 0.83W, respectively [1]. However, this study also shows that, if the radio is put in sleeping state, the consumption falls to 0.13W. Those values indicate that to have an effective decrease in energy consumption, the radio should be put in a sleeping state during certain periods of the operation of the network. This work was partially supported by Capes, CNPq and FAPERJ

2 In this work, an algorithm, Extra, is presented to deal with the problem of coordination of node transition to a sleeping state in ad hoc networks. This algorithm allows each node in the network to make decisions in an autonomous way, based only on local information. Additionally, it can work together with any ad hoc routing protocol. This paper is organized as follows. Section 2 presents some related work. A detailed description of the Extra algorithm follows in Section 3. Section 4 presents the evaluation scenario we set for our simulations, and Section 5 the simulation results. Finally, in Section 6 we draw our conclusions and make suggestions for further work. 2 Related Work The idea of shutting down the radio to conserve energy in ad hoc networks was already explored by other authors. In the BECA algorithm [8], the decision to put a node to sleep is taken based on local informations. If the node does not have a message to transmit or if it does not listen any message it goes to the sleeping state. After a certain period it wakes up and listen to messages. If no messages are heard, it goes back to sleep. In the AFECA algorithm [8], each node uses an estimation of the number of neighbor nodes to increase/decrease the time it will stay in the sleeping state. With a increase in node density, more energy can be saved. GAF [9] uses geographical position information (e.g., produced by a GPS) to support energy conservation. Like AFECA and BECA, GAF is independent of the underlying routing protocol. In this algorithm, the area where the network is active is divided in a virtual mesh, composed by fixed squares. The nodes in each square change between the sleeping and active states, but guaranteeing that one of them is active, in order to maintain the network connectivity. The choice of which node will be active is made in a distributed way. In CEC [7] the authors try to avoid the dependence that GAF has on location information (a GPS does not work inside buildings). Experiments accomplished by the authors demonstrate that CEC has a better performance than GAF for low mobility scenarios. 3 The Extra Algorithm Our energy saving procedure is based on putting nodes in the sleeping state. The main characteristics of this energy saving mechanism, that makes it different from other proposals, are the the moment a node decides going to sleep and the duration of sleeping state. Because the decision is taken in a distributed way, some nodes enter in the sleeping state for a known period, while others will stay active. In highly populated networks, there is a greater chance that few nodes stay active, maintaining network connectivity, while other nodes save energy. When the density decreases, some cautions are taken to reduce the disconnection probability.

3 3.1 Algorithm operation A node can be in one of three states: Active, Listening or Sleeping, as indicated in Figure 1. Each node starts operation sending a hello message, reporting to its neighbors that it is active. In this state a node has its radio on and will stay active for Ta seconds. If during this period it receives a message or a route response, the timer is restarted. After T a seconds without a message being received, the node runs the decision procedure that will determine, with probability P, whether it will go to sleep or it will stay active. Fig. 1. Node state transition in Extra If the decision is to stay active, another cycle of Ta seconds is started; otherwise, the node starts the energy saving procedure, sending to its neighbors a message informing that it will move to the Sleeping state. This allows each node to estimate its number of neighbors and how many are in the Active state. A node stays in the Sleeping state for Td seconds. After that, it changes to the Listening state. In this situation, the node listens to verify if either there is a message for him, or a broadcast message. If this is true, the node should go to the Active state. In case there are no messages during Te seconds, the node goes back to the Sleeping state. This cycle is repeated N times (see Figure 2) and then the decision procedure is again executed. At any moment, if an application running in a node needs to send a message, this node should become active. When an active node needs to route a message to a neighbor the transmission should be made only when this neighbor is in the Listening or Active states. If the neighbor is in the Sleeping state the transmission should be delayed until it changes state. In order to control the neighbors state, each node has a cache mechanism, updated when a message is received. A new neighbor is included in the cache when the node receives a route request or a hello message. The information on all neighboring nodes are always saved with a time stamp. If, after a time interval, no new message is received from a certain neighbor, its cache entry is considered obsolete.

4 Fig. 2. Example of temporal evolution 3.2 Obtaining Probability P A node starts an energy saving period with probability P, and this value can change during run time. This is done to adjust the energy saving mechanism to the network conditions. In our implementation, only information regarding the connectivity and energy levels are considered. Simply put, P should be high when the energy level is low and also when the number of neighbors is high. Based on this statement, some heuristics were developed. Heuristic 1 (h1): its characteristic is to try to preserve network connectivity, despite the energy level of the battery. Thus, a greater P is attributed to the nodes in high density areas. P is defined by the product of the number of neighbors by a factor K. However, to avoid that all the nodes in a very populated region go asleep at the same time, P is limited by a constant L: P(h1) = K neighbors number() if P(h1) > L then P(h1) = L end if where neighbors number() returns the amount of neighbors of a node. In our simulations K and L were set to 0.1 and 0.9, respectively. Heuristic 2 (h2): its characteristic is also to try to preserve network connectivity. In this case, P will be the relation of the active neighbors to the total number of neighbors. Heuristic 3 (h3): the main objective of this heuristics is to save energy when battery is at a low level, in spite of the connectivity: P(h3) = 1 current energy()/initial energy() where current energy() is returns the remaining energy of the battery and initial energy() returns its initial value. Heuristics 4 (h4): this heuristics is a combination of the heuristics 1 and 3. It tries to save energy balancing network connectivity and energy conservation at the node:

5 if (P(h1) ref) and (P(h3) ref) then P(h4) = max(p(h1),p(h3)) else if (P(h1) ref) and (P(h3) < ref) then P(h4) = P(h1) else if (P(h1) < ref) and (P(h3) ref) then P(h4) = P(h3) else if (P(h1) < ref)and(p(h3) < ref) then P(h4) = min(p(h1),p(h3)) end if where P(h1) and P(h3) are the values of P returned by the heuristics 1, and 3, respectively, and ref is an adjustable parameter, whose value in the simulations was set to 0.5. Heuristics 5 (h5): has a similar objective to heuristics 4. It is the combination of the heuristics 2 and 3. It tries to save energy when it is at a low level, yet trying to maintain network connectivity: if (P(h2) ref) and (P(h3) ref) then P(h5) = max(p(h2),p(h3)) else if (P(h2) ref) and (P(h3) < ref) then P(h5) = P(h2) else if (P(h2) < ref) and (P(h3) ref) then P(h5) = P(h3) else if (P(h2) < ref) and (P(h3) < ref) then P(h5) = min(p(h2),p(h3)) end if 4 Evaluation Scenario The Extra algorithm was evaluated through simulation. To assess its performance, it was compared to AODV, as the reference protocol, and to GAF running over AODV. To validate the simulation model, each metric was measured in 10 different scenarios, and the average value is presented with a 95% confidence interval. The simulator used in the tests was the Network Simulator - ns-2, version 2.26 [3]. To create the simulation scenario it was used the BonnMotion software, version 1.1 [4]. In all simulations the values attributed to N, Ta, Te and Td were, respectively, 15, 4s, 0.05s and 0.5s. The simulation scenario assumed 60 nodes, moving randomly (random way-point model) in a 1200m by 600m area. The nodes speeds were set to 0m/s (a static network used as reference), 1m/s (a person walking), and 10m/s (a vehicle in an urban environment). Also, four static nodes were placed near the edges of the referred area, two acting as source and two as sink of traffic. The reason for that was to evaluate the message delay and drop rates in a more coherent way, since intermediate nodes will serve exclusively for message routing.

6 The simulation time was 900 seconds, to permit a comparison with measures done for the GAF algorithm in [9]. In all scenarios, we used pause times of 0, 30, 60, 120, 300, 600 and 900 seconds. The radio range was set to 250m, and the propagation model was the two-ray-ground. For the traffic model we choose two sources and two sinks implemented in fixed nodes, with CBR traffic over UDP transport. The transmission rates were 1pkt/s, 10pkts/s and 20 pkts/s, producing an aggregate of 2, 20 and 40pkts/s, respectively. Message size was set to 512 bytes. The energy model chosen is based on the measures from [6] of the WaveLAN 2 Mb/s board, that is 1.6W for transmission, 1.2W for reception, 1.0W in idle mode and 0.025W in the sleeping state. To allow for a comparison, those values are the same used for GAF evaluation in [9]. As the energy consumption in the sleeping state is much smaller than in the other cases, and to simplify the algorithm implementation in ns-2, we considered that when a node is sleeping it is shutdown. This simplification was done for all algorithms. The initial energy attributed to each node was 500J, enough to maintain the network working for about 450s with the AODV protocol. In this case, since the energy consumption in the idle state is important, we could observe that almost all nodes will run out of energy by that time, independently of the traffic forwarded. Given the simulation time is set to 900s, we can thus observe the behavior of our energy saving mechanism. As it is not interesting to evaluate the network behavior when a source/sink shut down by lack of energy, we set an infinite energy level for those nodes, guaranteeing that they will always transmit during the simulation period. 5 Simulation Results The first experiment shows the extension of network lifetime obtained by Extra, when compared to GAF and AODV (see Figure 3). In this experiment it was assumed a pause time of 0s, maximum node speeds of 1m/s and a rate of 10 pkts/s was maintained for each data source. In this figure we can clearly see the benefit obtained by our algorithm. Given Extra has a probabilistic basis, we analysed the risk of network disconnection due to all surviving nodes deciding to sleep simultaneously. Figure 4 shows, for heuristics 4, the percentage of operative nodes, that is, nodes either active or listening, considering at any moment just the surviving nodes. In this experiment, pause time, transmission rate and node speed were equal to 0s, 1pkt/s and 1m/s, respectively. As it can be seen, roughly half of the surviving nodes are active or listening during the whole period. This shows that the probability of complete disconnection is negligible. Figures 5 and 6 show the effect of GAF and Extra on network lifetime, for several transmission rate and node speed values. These graphics present the percentage of surviving nodes at 900s. For lack of space, in all those cases the results are shown just for heuristics 4. In each figure, pause time is equal to 0s and, when not used as a control parameter, transmission rate and node speed

7 Fig. 3. Lifetime: AODV, GAF and Extra Fig. 4. Percentage of operative nodes were kept constant and equal to 1pkt/s and 1m/s, respectively. For the full range of transmission rates experimented, Extra overcomes GAF. In particular, Extra shows a far better performance for low rates (less than10pkts/s). For the class of applications we proposed Extra (i.e., low mobility, as in the case of rescue teams), our algorithm shows better performance and an almost steady behavior as speed is varied. We also studied the loss of packets (drop rate) using the same control parameters as in the previous experiment. In all three cases Extra and GAF produced close results, with a slight advantage for GAF. Fig. 5. Surviving nodes versus trans. rate Fig. 6. Surviving nodes versus speed 6 Conclusions and Future Works This work presents a new algorithm, Extra, for energy conservation in ad hoc wireless networks. As well as in several other algorithms, its objective is to

8 maximize the lifetime of the network, but yet trying to maintain the network connectivity. Its operation is controlled by a function that defines when the node should enter in a sleeping state. The sleeping time is constant, and the entrance in this state happens with a certain probability. This probability is calculated during execution time through an heuristic method. Five different heuristics were tested, all of them with very simple structure. As stated, all heuristics use only information obtained locally. As a proposal for a future works, solutions where the neighboring nodes exchange information about their current energy level and neighbors density could be explored. Obviously, this would cause an increase in the number of control messages, and implies in additional energy consumption. However, it is possible that this negative effect will not be enough to degrade the performance. Based on the simulation results, we believe that the selection of a small number of parameters and the choice of simple heuristics could produce an important impact on the extension of ad hoc networks lifetime, after all entia non sunt multiplicanda praeter necessitatem. References [1] B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris. Span: An energy-efficient coordination algorithm for topology maintenance in ad hoc wireless networks. Wireless Networks, 8(5): , [2] D.B. Johnson and D.A. Maltz. Dynamic source routing in ad hoc wireless networks. In T. Imielinski and H. Korth, editors, Mobile Computing, pages Kluwer Academic Publishers, January [3] ns-2. The network simulator - ns-2, [4] University of Bonn. Bonnmotion: A mobility scenario generation and analysis tool. [5] C.E. Perkins and E.M. Royer. Ad hoc on-demand distance vector routing. In IEEE Workshop on Mobile Computing Systems and Applications, pages , New Orleans, LA, USA, February [6] M. Stemm and R.H. Katz. Measuring and reducing energy consumption of network interfaces in hand-held devices. IEICE Transactions on Communications, E80- B: , [7] Y. Xu, S. Bien, Y. Mori, J. Heidemann, and D. Estrin. Topology control protocols to conserve energy in wireless ad hoc networks. Technical Report 6, Center for Embedded Networked Sensing, Los Angeles, USA, January [8] Y. Xu, J. Heidemann, and D. Estrin. Adaptive energy-conserving routing for multihop ad hoc networks. Technical Report 527, USC/ISI, Los Angeles, USA, October [9] Y. Xu, J. Heidemann, and D. Estrin. Geography-informed energy conservation for ad hoc routing. In ACM/IEEE Int. Conf. on Mobile Computing and Networking, pages 70 84, Rome, Italy, July 2001.