Traffic Analysis & Modeling in Wireless Sensor Networks and Their Applications on Network Optimization and Anomaly Detection 1,2

of 19

Please download to get full document.

View again

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.
PDF
19 pages
0 downs
3 views
Share
Description
Traffic Analysis & Modeling in Wireless Sensor Networks and Their Applications on Network Optimization and Anomaly Detection 1,2
Tags
Transcript
  Network Protocols and Algorithms ISSN 1943-3581 2010, Vol. 2, No. 1 74 Traffic Analysis & Modeling in Wireless Sensor  Networks and Their Applications on Network Optimization and Anomaly Detection 1,2   Qinghua Wang 3  Dept. of Information Technology and Media, Mid Sweden University SE-85170 Sundsvall, Sweden Tel: +46-60-148914 E-mail: qinghua.wang@ieee.org Abstract Wireless sensor network (WSN) has emerged as a promising technology thanks to the recent advances in electronics, networking, and information technologies. However, there is still a great deal of additional research required before it finally becomes a mature technology. This article concentrates on three factors which are holding back the development of WSNs. Firstly, there is a lack of traffic analysis & modeling for WSNs. Secondly, network optimization for WSNs needs more investigation. Thirdly, the development of anomaly detection techniques for WSNs remains a seldom touched area. Among these three factors, the understanding regarding the traffic dynamics within WSNs provide a basis for further works on network optimization and anomaly detection for WSNs. Keywords: Anomaly detection, network optimization, traffic analysis, traffic modeling, wireless sensor network. 1  This work has been based on the author’s PhD dissertation [1]. 2  This work was partially carried out during the tenure of an ERCIM “Alain Bensoussan” Fellowship Programme. 3  Current address: Department of Electronics and Telecommunications, Norwegian University of Science and Technology, Trondheim N-7491, Norway. Tel: +47-73594331.  Network Protocols and Algorithms ISSN 1943-3581 2010, Vol. 2, No. 1 75 1. Introduction Wireless sensor network (WSN) has emerged as a promising technology because of the recent advances in electronics, networking, and information processing. The WSN research was initially driven by military applications such as battlefield surveillance and enemy tracking. Now, many civil applications of WSN have also been proposed, which include habitat monitoring, environmental observation and forecasting systems, health monitoring, etc. In these applications, many low power and inexpensive sensor nodes are deployed in a vast space to cooperate as a network. Although WSN is a promising technology which can be used in many applications, there are still a few obstacles to overcome before it finally becomes a mature technology. One of the key obstacles is the energy constraint suffered by the most inexpensive sensor nodes, where batteries are the main source of power supply. Given this obstacle cannot be removed in the near future, optimizing the design of WSNs thus the minimum energy will be consumed is very important. In WSNs, communication is believed to dominate the energy consumption [2]. Energy expenditure is less for sensing and computation. The energy cost of transmitting 1 Kb a distance of 100 meters is approximately the same as that for the execution of 3 million instructions by using a general-purpose processor [3]. Thus, minimizing the energy consumption due to communication is the key for the relief of the energy constraint in WSNs. Currently, the knowledge about the communication in WSNs is still partial and vague, especially for traffic characteristics and communication patterns. Obviously, the knowledge about the traffic characteristics and communication patterns can aid in the understanding of the energy consumption and its distribution in WSNs. Thus, the investigation of traffic characteristics and communication patterns is a good starting point in the search for more energy-efficient WSNs. Following on from this it will be possible to propose new solutions for the design of WSNs in order to optimize the energy consumption. Another concern for WSN technology involves security. WSNs will not be successfully deployed if the security issue is not addressed adequately. Security becomes more important  because WSNs are usually used for very critical applications. Furthermore, WSNs are very vulnerable and thus attractive to malicious attacks because of their cheap prices, human-unattended deployment and the nature of wireless communication. The existing solutions to the security in WSNs include using key management and authentication [4]. However, these preventive mechanisms cannot deter all possible attacks (e.g. insider attacks  possessing the key). Actually, malicious attacks may exhibit anomalous behaviours in WSNs. With regard to communication, malicious attacks can trigger arbitrary communications, while a normal communication must follow protocol specifications and application scenarios. Thus, it should be interesting to investigate the possibility of detecting malicious attacks by identifying the anomalies exhibited within the WSNs' communication traffic. Because sensor nodes are cheap devices and they can be deployed in harsh environments (e.g. battlefield, forest), they are prone to fail either by themselves or by means of others (e.g.  Network Protocols and Algorithms ISSN 1943-3581 2010, Vol. 2, No. 1 76 enemies, animals). Further, it is also common for battery-supported sensor nodes to fail  because of energy exhaustion. To provide efficient maintenance for WSNs, those performing this maintenance require instant notifications about the sensor node failures. Because a failed sensor node cannot maintain efficient communication with the other nodes, sensor node failures have the possibility to be instantly noticed by observing the degraded or lost communication in relation to the failed nodes. This strategy has a similarity with the detection of traffic anomalies caused by malicious attacks. Both of them require comprehensive knowledge about the communication traffic before they can identify any traffic anomaly. The aim of this article is to investigate the communication traffic dynamics and patterns in WSNs and find their applications with reference to network optimization and network anomaly detection. The applications of WSNs are abundant. Because the communication traffic in WSNs is very dependent on the application scenario, only those selected typical WSN scenarios (e.g. surveillance, target tracking) will be investigated. Additionally different types of communication traffic exist, including data traffic, routing discovery traffic, link layer feedback and hello message, etc. This article mainly focuses on data traffic and there is a limited involvement of other traffic types. In the following, the survey of the works in the field of traffic analysis & modeling, in the field of network optimization and in the field of network anomaly detection will be  presented separately. However, particular emphasis will be put on the relationships between traffic analysis & modeling and network optimization, and between traffic analysis & modeling and network anomaly detection throughout the presentation. 2. Traffic Analysis & Modeling for WSNs WSNs consist of a large number of tiny and cheap sensor nodes that cooperatively sense a physical phenomenon. Existing research results and products have provided the possibility to build effective WSNs for many applications. If the traffic features inside WSNs were better understood then the WSNs could be made to be even more effective. For example, better routing protocols and sensor deployment strategy could be designed if the traffic burden among the sensors was better understood. Better fault and security management could be applied if normal and abnormal traffic could be kept apart according to traffic features. The traffic dynamics for different types of traditional networks, both wired and wireless, have been investigated in the literature. However, the specialty of WSNs makes a reinvestigation of traffic dynamics necessary. Constructing accurate and analytically tractable models for sensor network traffic will provide a basis for future work on network design, optimization and security. Unfortunately, at the time that this article was written, research regarding traffic modeling and analysis in WSNs was still rather limited. The few studies that do exist include works focusing on data traffic arrival process, sequence relations among general kinds of packets, and data traffic load distribution. 2.1 Data Traffic Arrival Process Because the data traffic dynamics in different WSN scenarios are quite different, the data  Network Protocols and Algorithms ISSN 1943-3581 2010, Vol. 2, No. 1 77 traffic modeling and analysis in WSNs will be quite application dependent. In [5], it is suggested that WSN applications can be categorized as event-driven  or  periodic data generation . For periodic data generation scenarios, constant bit rate (CBR) can be used to model the data traffic arrival process when the bit rate is constant [6]. When the bit rate is variable, a Poisson process can be used to model the data traffic arrival process as long as the data traffic is not bursty [7]. For event-driven scenarios such as target detection  and target tracking , bursty traffic can arise from any corner of the sensing area if an event is detected by the local sensors. A Poisson process has also been used to model the traffic arrival process in an event-driven WSN [8]. However, there is no solid ground to support the use of a Poisson  process in this case. Actually, the widely used Poisson processes are quite limited in their  burstiness [9, 10]. Instead of using Poisson processes, the author of this article proposes to use an ON/OFF model (see Figure 1) to capture the burst phenomenon in the source data traffic of an event-driven WSN [11]. Further, the distributions of ON/OFF periods are found to follow the generalized Pareto distribution in his considered WSN scenario. Ref. [12] studies a different WSN scenario - a mobile sensor network (MSN). In an MSN, the node mobility introduces new dynamics to network traffic. In [12], the authors find that the mobility variability of humans (in this case, sensor nodes are attached to humans) and the spatial correlation of the collected information lead to the pseudo-LRD (i.e. long range dependent) traffic, which exhibits characteristics significantly different to that of Markovian traffic. Figure 1: ON/OFF state transition diagram 2.2 Sequence Relations among General Kinds of Packets Sequence relations exist in some kinds of packets. For example, a Routing Reply message always comes after a Routing Request message and that is specified by any ordinary routing protocol. In [13], the authors propose to use a finite state machine (FSM) to specify correct routing behavior for the ad hoc on demand distance vector (AODV) routing [14]. The rationale behind this is that the AODV protocol has specified the sequence relations among different kinds of routing messages and such sequence relations can be depicted by an FSM. The authors in [15] also use FSM to model the correct routing behavior for the dynamic source routing (DSR) [16]. Because the routing protocols AODV and DSR have clearly specified the routing operations, the sequence relations among different kinds of routing OFF ON  No packet arrives A new packet arrives / restart ON timer ON timer expires A new packet arrives / start ON timer   Network Protocols and Algorithms ISSN 1943-3581 2010, Vol. 2, No. 1 78  packets can be manually abstracted into an FSM. In both [13] and [15], the authors have used their FSMs to validate real-time routing behaviors and detect possible malicious attacks. In addition to that the sequence relations among some special kinds of packets (e.g. routing messages) are possible to be specified according to protocol specifications, the author of this article suggests that the sequence relations among general kinds of packets can also be learned automatically by on-line training. In [17], the authors firstly classify the arriving  packets according to their attributes (e.g. packet type, addresses) and then map the packet arriving sequence to an infinite character string. Afterwards, the on-line learning of the packet sequence relations are conducted by extracting every unique character substring encountered during the window-based scanning process. The learned packet sequence relations can be used to build the normal traffic profile for the node of interest in a static WSN. In a dynamic WSN in which some of the nodes are mobile, the traffic profile learned in this manner will evolve quickly over time and will thus be less meaningful. 2.3 Data Traffic Load Distribution In a WSN, the data traffic load is not evenly distributed over the nodes. For example, the sensors which are one hop away from the sink relay the entire network's data traffic. This imbalanced data traffic load distribution can degrade the network's lifetime and functionality. Hence, efforts have been devoted to characterizing the data traffic load distribution in WSNs. Ref. [18] proposes an analytical analysis on the data traffic load distribution over a randomly deployed linear WSN. It has been shown that the data traffic load over a node increases the closer it is to the sink, however, a reduction in the data traffic load is expected for sensors that are very close to the sink. In [19, 20], data traffic load is formularized as a function of the distance to the sink in dense planar WSNs. In a similar manner to that in a linear WSN, the data traffic load over a node in planar WSNs also increases as the node moves closer to the sink. For a symmetric sensor network (i.e. all nodes of the same distance from the center of the network are similar) with nodes evenly distributed in the sensing field, the author of this article concludes that the expected data traffic load over a node is in direct proportion to the network radius, in inverse proportion to the mean routing hop length, and independent of the node density [20]. Because the distribution of data traffic load is closely related to the distribution of energy consumption and the latter has a significant impact on the performance of WSNs, the research results concerning the distribution of data traffic load can be used to optimize the  performance of WSNs. For example, the author of this article has proposed an optimal energy allocation scheme for WSNs based on the understanding of the data traffic related energy consumption in the network [21]. 3. Network Optimization for WSNs There are many network optimization problems to be solved in WSNs, such as rate control, flow control, congestion control, medium access control, queue management, power control and topology control, etc. [22]. It is difficult to provide a complete overview in relation to all issues relating to network optimization in WSNs. However, it is worthwhile,
Related Search
Advertisements
Advertisements
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