Medicine, Science & Technology

A Routing Scheme for Congestion Avoidance in Wireless Sensor Networks

Description
A Routing Scheme for Congestion Avoidance in Wireless Sensor Networks
Published
of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
     Abstract —   Congestion has an utterly detrimental influence in the performance of Wireless Sensor Networks (WSNs). Many of the causes of congestion in WSNs are different than the causes of congestion in wired networks. Hence, new research has to be developed for the avoidance and control of congestion in WSNs. This article proposes a scheme which aims to both forecast imminent congestion and prevent its further diffusion in WSNs. The prediction of forthcoming congestion is based on the computation of a suitably built cost function, which considers a set of metrics related with the popularity of the alternate nodes and routing paths. The prevention of congestion diffusion is carried out through multi hop/path routing, by utilizing nodes and routes which are less probable to be congested. Simulation tests performed evaluate the efficacy of the proposed scheme.  Index Terms — congestion avoidance, congestion control, multi hop/path routing, wireless sensor networks. I.   I NTRODUCTION  SNs typically consist of a set of many sensor nodes which are deployed over wide areas and transmit their data to a sink node, referred as the base station. Since the distance between a source node and another destination sensor node or the base station may exceed the range of the transmission ability of the sender, then relaying is required via intermediary sensor nodes. Therefore, a sensor node apart from gathering, processing and transmitting the data it senses, it may also have to forward data it receives from other sensor nodes [1-3]. Thus, the operation of a WSN is interdependently correlated with the transmission of great quantities of data. One of the most common problems that the management of the data traffic within a WSN faces is that of congestion. Congestion occurs when current traffic load exceeds D. Kandris is with the Department of Electronics, Technological Educational Institute (T.E.I.) of Athens, 12210, Athens, Greece (phone: +306972256162; e-mail: dkandris@teiath.gr). D. J. Vergados is with the School of Electrical and Computer Engineering, National Technical University of Athens, 15773, Athens, Greece (email: djvergad@telecom.ntua.gr). D. D. Vergados is with the Department of Informatics, University of Piraeus, 18534, Piraeus, Greece (email: vergados@unipi.gr). A. Tzes is with the Department of Electrical and Computer Engineering, University of Patras, 26500, Rio, Greece (email:tzes@ece.upatras.gr). available transmission ability at any point in the network. Congestion, similarly to what happens to wired networks, has an absolutely harmful influence in the performance of WSNs too. However, many of the causes of congestion in WSNs are different than the causes of congestion in wired networks. Consequently, new research work has to be carried out. This article proposes a scheme which aims to both predict upcoming congestion and prevent its further diffusion. The forecast of imminent congestion is based on the calculation of an appropriately built cost function, which takes into consideration a set of metrics related with the popularity of the alternate nodes and routing paths. The prevention of congestion diffusion is performed through multi hop/path routing, by utilizing nodes and routes which are less probable to be congested. The remainder of this article is organized as follows. In Section II, the reader is introduced to the causes and effects of congestion in WSNs. In Section III, an overview of existing congestion control and congestion avoidance schemes is presented. In Section IV, the proposed congestion handling scheme is described. The performance evaluation of the proposed scheme takes place in Section V, through the description of simulation results. Finally, section VI concludes the article. II.   C ONGESTION   IN WSN S  The data traffic in a WSN may be sorted out in two types. The first of them is the so called downstream   traffic , which is a one-to-many multicast communication, directed from the sink to the sensor nodes. The other kind of traffic is the so called upstream traffic , which is a many-to-one communication from sensor nodes to the sink. Due to the convergent nature of upstream traffic, congestion is more probable to appear in the upstream direction [4]. Actually, in WSN applications, the volume of traffic load in upstream traffic may have significant fluctuations caused by the existing demands. More precisely, the traffic may range from simple periodical or query-based reports to sudden outbursts of event-based data flow triggered in response to sensed incidents. In most cases, such event-based increases of traffic load amplify the probability of congestion occurrence. A Routing Scheme for Congestion Avoidance in Wireless Sensor Networks Dionisis Kandris, Dimitrios J. Vergados, Dimitrios D. Vergados, and Anthony Tzes W 6th annual IEEE Conference on Automation Science andEngineeringMarriott Eaton Centre HotelToronto, Ontario, Canada, August 21-24, 2010 MoB3.2 978-1-4244-5448-8/10/$26.00 ©2010 IEEE497   Additionally, probable changes inthe communication rates adopted within a WSN may cause congestion to previously uncongested parts of the network. Moreover, even under a certain periodic traffic pattern, congestion may still appear. This is due to the interference caused by simultaneous transmissions of neighboring nodes. Furthermore, congestion may be also inflicted in a WSN, due to alterations of the node population due to either addition of new sensor nodes or depletion of existing ones. Finally, congestion may be caused due to variations of the quality of wireless communication channels [5]. Congestion has many significant consequences. The first of them is the expansion of buffer overflows which leads to the raise of packet losses and thus the gradual reduction of the network throughput. Another after-effect is the increase of energy dissipation due to the increase of retransmissions that the data source nodes are obliged to perform in order to send their messages to the receiving nodes which are congested. Moreover, along with the ongoing network collapse, the network QoS diminishes, and the tasks of reliable data transmission and prompt event detection become practically infeasible. III.   R ELATED W ORK  Nowadays, a lot of research is being carried out on the scientific fields of congestion control and congestion avoidance in WSNs. The field of congestion control includes the methodologies which aim to perform in a systematic way the network recuperation from congestion consequences. On the other hand, the field of congestion avoidance focuses on techniques which intend to perceive congestion at its early stages and prevent its further diffusion. These techniques may be classified into four categories, which namely are: data aggregation, flow control, sink mobility and multi hop/path routing techniques. This section presents the main features of these four categories.  Data aggregation  techniques make use of the correlation existing among sensed data in order to decrease the quantity of data transmitted from the nodes to the base station and hence prevent congestion. More specifically, in applications where there is low variation over time or space among consecutive data measurements made by the sensor nodes, temporal aggregation [6] or spatial aggregation schemes may be applied [7]. An even greater reduction of the traffic load can be achieved by the utilization of tempo-spatial aggregation in the way that it is used in energy efficient hierarchical routing protocols [8, 9]. Normally, data aggregation when compared with traditional forwarding of data is associated with an increase of both the processing cost and the probability of losing important pieces of information. Thus, when using data aggregation techniques, the tradeoff between the degree of data aggregation, the processing cost and the level of information integrity has to be investigated [10]. The basic idea of   flow control  techniques is the attempt to control the amount of data transmitted in order to avoid congestion. There are examples of this type of techniques which make use of back pressure mechanisms which are activated as soon as congestion is detected in order to make upstream nodes to reduce their transmission rates [11]. Other schemes are based on the assumption that client nodes have complete information about the buffer status of their parent node. In [12] an interference-aware rate control mechanism which adapts the transmission rate according to the size of incipient congestion is proposed. The detection of forthcoming congestion is performed by monitoring the average queue length of every single node and sharing this information among all neighbors. Similarly, in [13] in case of congestion the downstream nodes either reduce the data that they currently forward to an upstream node or they switch to other upstream nodes. Another idea proposed is the assign a different priority to each node according to its traffic load in order to let more access to the transmission media to the nodes having heavy data traffic [4]. Generally, flow control techniques are effective in restraining congestion in WSNs. However, the use of traffic limitations and back pressure mechanisms lead to data loss. Sink mobility  schemes are based on the use of a mobile base station which moves close to the transmitting sensor nodes. Although the use of these schemes in WSNs has initially been introduced in order to enable load balancing and consequently energy efficiency, it may also enable the mitigation of congestion with the disadvantage of increased data latency and packet loss. In sink mobility schemes the mobility pattern may be either random [14], fixed [15-17] or controlled [18]. According to their type, sink mobility techniques suffer from certain drawbacks such as the occurrence of extensive latency and data loss and the availability of complicated and costly infrastructures. In general, the basic idea of multipath routing models is the discovery of multiple paths from source to destination in order to route around failed nodes or faulty links [19, 20]. Similarly, multi hop/path routing  congestion avoidance schemes focus on the discovery of alternate routing paths for the forwarding of data, in order to share the data traffic and eliminate congestion from the network. In [21] a typical example of a multi hop/path routing scheme is presented, which aims at reallocating traffic through alternate routes and providing network load balancing. Similarly, [22] tries to detect early congestion based on the knowledge of both the queue sizes of the buffers of the upstream nodes along with the relationship between the numbers of downstream and upstream nodes. Multi hop/path routing techniques are generally considered to perform very well in mitigating congestion and avoiding data losses. However, in many cases the mitigation of congestion in such techniques is deteriorated at the nodes which are located at the vicinity of 978-1-4244-5448-8/10/$26.00 ©2010 IEEE498   the base station. IV.   P ROPOSED S CHEME D ESCRIPTION  Many of the aforementioned schemes are associated with considerable data losses. Some others have low cost effectiveness. This article proposes a novel scheme, which aims at two targets. The first of them is to propose a systematic way for the identification of upcoming congestion. This is methodized through the computation of a suitably formulated cost function which evaluates the influence of various factors that affect the congestion occurrence. The second research target of this article is the introduction of a multi hop/path routing scheme in order to avoid congestion through the evasion, during upstream routing, of the nodes which are more probable to cause congestion. More precisely, in a network consisting of stationary nodes, the network topology may be considered as an invariant factor which affects the probability of each individual node to receive more traffic. For instance, nodes located close to the base station are associated with a greater probability to be intermediate routers of data transmitted from more distant source nodes to the base station. Thus, these nodes due to their greater popularity are more probable, to receive higher traffic load and consequently become bottlenecks. In a similar way, nodes having less upstream neighbors than downstream neighbors are theoretically associated with a greater probability to receive incoming traffic load which has rates higher than the corresponding outgoing traffic and thus cause congestion. Finally, nodes which are located at positions closer to the areas of often incident occurrence have increased traffic load due to the many event-based reports they have to transmit to the base station. In order to evaluate the effect of the aforementioned factors over the likelihood of congestion occurrence, the cost that every individual node has, concerning three different metrics, is calculated. These metrics are: x   The distance existing between the node and the base station. x   The ratio of the total number of the neighbors of a node over the number of its upstream neighbors. x   The total time of the node participation in communication flows. The individual influence of every one of these three factors has to be investigated. In order to do so, for each of them a corresponding routing scheme is created, based on a corresponding cost function: 1)   The routing scheme which is based on the distance existing between the node and the base station, tries to reduce the traffic load by circumventing during routing, if possible, nodes located close to the base station. Thus the cost C  i  for a node i  according to this metric is: % Ü  L & Ü   : s ;  where  D i  denotes the distance between the node i and the base station. 2)   The routing scheme which is based on the ratio between the numbers of all neighboring nodes over upstream neighboring nodes tries to avoid routing via nodes having more downstream than upstream neighbors. The cost C  i  of a node i  according to this metric is equal to: % Ü  L 0 Ü 7 Ü   : t ;  where U i   denotes the number of upstream neighbors of node i  and  N  i  stands for the total of its neighboring nodes. 3)   The routing scheme which aims to take into consideration the frequency of incident occurrence, tries to circumvent during routing the nodes that are associated with an increased popularity of use. For this reason, the cost C  i  of a node i  according to this metric is represented as the total of use time T  i . Thus: (3) ii CT    The combined routing scheme proposed, joins the three routing metrics in order to obtain better results. The cost of each node is calculated as follows: Initially a weighting factor is calculated for every one of these three metrics by calculating the maximum value of that metric for the entire network. Additionally, another routing scheme which takes into consideration all three aforementioned metrics is introduced. It is based on the use the evaluation of the so called popularity cost which is calculated as follows: Let S   be the set of s  nodes constituting a network topology, V  mi  be the cost value of node i  by using metric m  and V  m  be the maximum cost value calculated by using metric m : 1  max{} (4) mmiis VV  dd   The weighting factor W  mi   of node i  as to metric m is: 1, 0, )0,(50 mmmim ifV V WiS ifV  -°•®°¯z  Then the Popularity cost PC  i of a node i  is calculated as: 978-1-4244-5448-8/10/$26.00 ©2010 IEEE499    13  () , (6) imimim PCWViS   •˜ ¦  V.   P ROPOSED S CHEME E VALUATION In order to evaluate the efficacy of the aforementioned metrics, extended simulation tests were performed by using a custom-built simulation tool. In all simulation tests performed the same square topology having each side equal to 90 m was used. This deterministic topology consists of 100 equidistant nodes and 1 base station located at the center of the network field. The communication range of a node is equal to 11 m. The distance between neighboring nodes, which are located at the same either row or column, is set to be 10 m. During simulation time incidents are generated at random instances and locations within the network field. Incidents are supposed to cover a cyclic area of gradually increasing radius, which takes values in the range from 7.5 m up to 50 m. At the occurrence of an incident, each node that is located within the incident area is supposed to transmit a burst of data of total size 1 Mb. The channel rate is supposed to be equal to 1 Mbps. The inter-arrival time between each burst generation was exponentially distributed. Each simulation test is completed as soon as 100 data flows are completed. In total, four different routing schemes were examined in simulation tests performed, i.e. the aforementioned in Section routing schemes corresponding to the three metrics and a combined one which takes into consideration all three metrics. The performance of every one of these four routing schemes was examined against the routing via the shortest path. Additionally, since the center of attention in this research work is the network layer, an ideal Medium Access Control protocol, that divides the channel fairly among the competing nodes was assumed to exist. Moreover, an ideal Transport Control Protocol is assumed, that instantly adapts the transmission rate to the available by the network throughput. The routing schemes mentioned above in Section IV, were initially compared, by evaluating the average time needed for end-to-end data transmission of bursts of data of size 1 Mb against the increase of the size of incident areas. The results are depicted in Fig. 1. As it can seen in Fig.1, in all routing schemes compared, the increase of the radius of incidents causes a corresponding increase of the average time needed for the end-to-end data transmission. This is absolutely normal due to the fact that as the area influenced by the presence of an incident gets larger, the number of nodes which sense this event increase too. In this way, the data traffic raises too thus causing more conflicts among data flows directed to the base station. However, all routing schemes proposed, have a shorter transmission delay than the routing through the shortest path. Moreover, the combined routing scheme outperforms all other routing schemes. Fig. 1. Graphical presentation of the average time needed for the end to end transmission of data bursts of 1 Mb size over the radius of incidents generated within the network field, when routing is performed according to a) the shortest path, b) the distance existing between the node and the base station, c) the ratio of the total number of the neighbors of a node over the number of its upstream neighbors, d) The total time of the node participation in communication flows or e) the combined use of the last three metrics. The percentage decrease of the average end-to-end transmission time achieved through the use of the individual aforementioned schemes when compared against the shortest path is depicted in Fig. 2, which illustrates more clearly the performance of all proposed routing schemes in comparison to the scheme which routs data through the shortest path. It is shown that all proposed schemes perform better by the increase of the incident area. Only when the smallest incident area is applied, the routing scheme which considers the ratio of the total of neighboring nodes over the number of upstream nodes performs worse than the routing via the shortest path. This is justified by the fact that for short incident areas, there are few nodes participating in the data routing process. Thus for low data traffic conditions routing via the shortest path is a quite good solution. However, by the increase of the data traffic transmission via the shortest path gets more and more delayed. Finally, the various routing schemes are compared in regard of the network throughput. More specifically, the variation of the overall network throughput against the increase of the size of incident areas is evaluated for all schemes in comparison. In Fig. 3 the corresponding results are illustrated. 57,51012,51517,52022,52527,53032,55 7,5 10 12,5 15 17,5 20    A  v  e  r  a  g  e   T  r  a  n  s  m   i  s  s   i  o  n   T   i  m  e   (  s  e  c   ) Radius of Incident (m) CombinedDistanceNeighborsTime of UseShortest Path 978-1-4244-5448-8/10/$26.00 ©2010 IEEE500    Fig. 2. Graphical presentation of the percentage decrease of the average time needed for the end to end reception of data bursts of 1 Mb size over the radius of incidents generated within the network field, when routing is performed according to a) the distance existing between the node and the base station, b) the ratio of the total number of the neighbors of a node over the number of its upstream  neighbors, c) The total time of the node participation in communication flows or d) the combined use of the last three metrics. As it can be seen in Fig. 3, the increase of incident area causes the corresponding gradual decrease of the overall network throughput. However, the rate of throughput reduction is higher when routing is performed through the shortest path, while the combined proposed routing scheme outperforms all other schemes. Fig. 3. Graphical presentation of the overall network throughput against the the size of incident areas time needed for the end to end reception of different burst sizes when routing is performed according to a) the shortest path, b) the distance existing between the node and the base station, c) the ratio of the total number of the neighbors of a node over the number of its upstream  neighbors, d) The total time of the node participation in communication flows or e) the combined use of the last three metrics. VI.   C ONCLUSIONS  This article focused on the management of congestion in WSNs. Related work in currently existent congestion avoidance and congestion control techniques was presented along with an overview of their corresponding advantages and weaknesses. Various factors that affect the probability of congestion occurrence were discussed. Finally, a novel scheme was proposed, which aims to both detect upcoming congestion by taking into consideration and evaluating these factors and prevent its further diffusion through multi hop/path routing. The efficacy of the proposed scheme was evaluated through simulation tests. R EFERENCES   [1]   I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “A survey on sensor networks,”  IEEE Communications Magazine , vol. 40, no 8, pp.102-114, 2002. [2]   I. F. Akyildiz, I.H. Kasimoglu, Wireless sensor and actor networks: Research challenges,  Ad Hoc Networks , vol. 2, no. 4, pp. 351-367, 2004. [3]   F. Zhao and L. Guibas, Wireless Sensor Networks , Elsevier, 2004, pp. 1-20. [4]   C., Wang, B. Li, K. Sohraby, M. Daneshmand, and Y. Hu, “Upstream congestion control in wireless sensor networks through cross-layer optimization,”  IEEE journal on selected areas in communications , vol. 25, no.4, 2007, pp. 786-795. [5]   B. Hull, K. Jamieson and H. Balakrishnan, “Mitigating Congestion in Wireless Sensor Networks,” in Proc.   SenSys'04, Baltimore, MD, USA, November 3-5, 2004. [6]   M. A. Sharaf, J. Beaver, A. Labrinidis, and P. K. Chrysanthis, “TiNA: a scheme for temporal coherency-aware in-network aggregation,”  in Proc.   3 rd   ACM International Workshop on Data Engineering for Wireless and Mobile Access MobiDe '03 , San Diego, CA, USA, September 19, 2003, pp. 69-76. [7]   L. Galluccio, A. T. Campbell, and S. Palazzo, “Concert: aggregation-based congestion control for sensor networks,”  in Proc.  3 rd  international conference on embedded networked sensor systems 2005, San Diego, CA, USA, pp. 274-275. [8]   A. Manjeshwar, and D.P. Agrawal, “TEEN: a routing protocol for enhanced efficiency in wireless sensor networks,”  in Proc.   15 th   International Parallel and Distributed Symposium , San Francisco, CA, USA, April 2001, pp. 2009–2015. [9]   D. Kandris, P. Tsioumas, A. Tzes, G. Nikolakopoulos, and D.D. Vergados, “Power conservation through energy efficient routing in wireless sensor networks,” Sensors , vol. 9, pp. 7320-7342, Sept. 2009. [10]   L. Galluccio, S. Palazzo and A.T. Campbell, “Efficient Data Aggregation in Wireless Sensor Networks: an Entropy-driven Analysis,”  in Proc. 19th IEEE International Symposium on Personal,  Indoor and Mobile Radio Communication PIMRC 2008  , September 15-18, 2008, Cannes, France, pp. 1– 6. [11]   C. Wan, S. B. Eisenman, A. T. Campbell, “CODA: congestion detection and avoidance in sensor networks,”  in Proc.   1 st   international conference on embedded networked sensor systems SenSys '03 , Los Angeles, CA, USA, November 05 - 07, 2003, pp. 266-279. [12]   S. Rangwala, R. Gummadi, R. Govindan, K. Psounis, “Interference Aware Fair Rate Control in Wireless Sensor Networks,” in Proc.   SIGCOMM’06  , September 11–15, 2006, Pisa, Italy. [13]   C. Shigang, N. Yang, “Congestion avoidance based on lightweight buffer management in sensor networks,”  IEEE Transactions on Parallel and Distributed Systems  vol. 17, pp. 934-946, 2006. [14]   I. Chatzigiannakis, A. Kinalis, and S. Nikoletseas, “Sink mobility protocols for data collection in wireless sensor networks,” in Proc.   -505101520255 7,5 10 12,5 15 17,5 20    D  e  c  r  e  a  s  e  o   f   T  r  a  n  s  m   i  s  s   i  o  n   T   i  m  e   (   %   ) Radius of Incident (m) CombinedDistanceNeighborsTime of Use 200004000060000800001000001200001400001600001800002000002200006 8 10 12 14 16 18 20    T   h  r  o  u  g   h  p  u  t   (   M   b  p  s   ) Radius of Incident (m) CombinedDistanceNeighborsTime of UseShortest Path 978-1-4244-5448-8/10/$26.00 ©2010 IEEE501
Search
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks