SEMINAR SENSOR

DATA MANAGEMENT IN SENSOR NETWORKS

CONTENTS

v INTRODUCTION

v SENSOR NETWORKS

v CHALLENGES

v CONCLUSION

v REFERENCES

INTRODUCTION

                                  Sensor networks have attracted a lot of attention lately and have been increasingly adopted in a wide range of applications and diverse environments, from healthcare and traffic management to weather forecasting and satellite imaging. A vast amount of small, inexpensive, energy-efficient, and reliable sensors with wireless networking capabilities is available worldwide increasing the number of sensor network deployments. Internet or to specific gateways and provide remote access for management and configuration issues. The adoption of IPv6 also provides a huge address space for networking purposes in order to address the large sensor networks on a global scale while concurrently leads to the rapid development of many useful applications. Thus, it is not unreasonable to expect that in the near future, many segments of the networking world will be covered with sensor networks accessible via the Internet. Archived and real time sensor data will be available worldwide, accessible using standard protocols and application programming interfaces (APIs).

                                   Nevertheless, as stated in too much attention has been placed on the networking of distributed sensing while too little on tools to manage, analyze, and understand the collected data. In order to be able to exploit the data collected from the sensor network deployments, to

Map it to a suitable representation scheme, to extract meaningful information from it and to increase interoperability and efficient cooperation among sensor nodes, we have to devise and apply appropriate techniques of data management. Towards this direction, data aggregation and processing have to be done in a way that renders it valuable to applications that receive stored or real time input and undertake specific actions. It is important to note that, special characteristics of sensor nodes, such as their resource constraints (low battery power, limited signal processing, limited computation and communication capabilities as well as small amount of memory) have to be considered while designing data management schemes.

                        Sensory data has to be collected and stored before being aggregated. Various techniques for data aggregation have been proposed in accordance with the type of the network and the imposed requirements. However, aggregated data is raw data that has little meaning by itself. Hence, it is crucial to interpret it according to information that is relevant to the deployed applications. This will increase interoperability among different types of sensors as well as provide contextual information essential for situational knowledge in the sensor network. Towards this direction, the Open Geospatial Consortium (OGC) recently established the Sensor Web Enablement (SWE) initiative to address this aim by developing a suite of sensor related specifications, data models and Web services that will enable accessibility to and controllability of such data and services via the Web. The Sensor Web is a special type of Web-centric information infrastructure for collecting, modeling, storing, retrieving, sharing, manipulating, analyzing, and visualizing information about sensors and sensor observations of phenomena.

DATA MANAGEMENT IN SENSOR NETWORKS

                             The increasing availability of small-size sensor devices during the last few years and the large amount of data that they generate has led to the necessity for more efficient methods regarding data management. In this chapter, we review the techniques that are being used for data gathering and information management in sensor networks and the advantages that are provided through the proliferation of Semantic Web technologies. We present the current trends in the field of data management in sensor networks and propose a three-layer flexible architecture which intends to help developers as well as end users to take advantage of the full potential that modern sensor networks can offer. This architecture deals with issues regarding data aggregation, data enrichment and finally, data management and querying using Semantic Web technologies. Semantics are used in order to extract meaningful information from the sensor’s raw data and thus facilitate smart applications development over large-scale sensor networks.   

SENSOR NETWORKS

Sensor Nodes: Functionality and Characteristics

                               A sensor node, also known as “mote”, was an idea introduced by the Smart Dust project in early 00's (2001). Smart Dust was a promising research project that first studied and supported the design of autonomous sensing and communication micro-computing devices of size as small as a cubic millimeter (or the size of a “dust particle”). In other words, this project acted as the cornerstone for the development of today’s wireless sensor networks. The key functionality of a modern sensor node, in addition to sensory data gathering, is the partial processing and transmission of the collected data to the neighbouring nodes or to some central facility. A modern node could be considered as a microscopic computer embedding all the units required for sensing, processing, communicating and storing sensory information, as well as power supply units able to support such operations. The most important units that are present in a sensor node are the following:

• The Processing Unit, that is responsible not only for processing the collected data, but also for orchestrating the cooperation and synchronization of all other mote's units towards realizing the promised functionality. Its operation is most often supported by on-chip memory modules.

• The Communication Unit, also known as transceiver that enables motes to communicate with each other for disseminating the gathered sensory data and aggregating them in the sink nodes (nodes with usually higher hardware

specifications than simple sensor nodes). The two most popular technologies considered here are either the Radio Frequency (RF) one, where the unlicensed industrial, scientific and medical (ISM) spectrum band is worldwide and freely usable by anyone, or the Optical or Infrared (IR) one, where line-of-sight between communicating nodes is highly required – making communication extremely sensitive to the atmospheric conditions.

• The Power Supply Unit, that provides power for the operation of such tiny devices. A typical power source does not exceed the 0.5Ah under a voltage of 1.2V and is most commonly a battery or a capacitor. While operations like data sensing and processing consume some power, the communication between neighbouring nodes is proved to be the most energy-consuming task.

• The Sensor Unit, that is responsible for sensing the environment and measuring physical data. Sensors are sensitive electronic circuits turning the analog sensed signals into digital ones by using Analog-to-Digital converters. There is a large variety of sensors available today with the most popular of them being able to sense sounds, light, speed, acceleration, distance, position, angle, pressure, temperature, proximity, electric or magnetic fields, chemicals, and even weather-related signals. Such units must be able to provide the accuracy the supported application demands, while consuming the lowest possible energy. Modern sensor nodes are required to be inexpensive, multifunctional, cooperative, microscopic, as well as able to cope efficiently with low power supplies and computational capacity.

Sensor Networks Topologies

                            When a number of sensor nodes are clustered together, a special type of autonomic and power

efficient network is formed, a so-called Wireless Sensor Network (WSN). WSNs are mainly consisting of the Sensor Nodes, the Sink Nodes that aggregate the measured data from a number of Sensor Nodes and the Gateway Nodes that interconnect the Sink Nodes with the network infrastructure (e.g. Internet) and route the traffic to proper destinations. There are cases where the Sink Nodes have embedded network interfaces for data forwarding and thus

coincide with the Gateway Nodes. Regarding the topology of the sensor network, it may form either a single-hop network where each Sensor Node sends directly the data to the Sink Node through a star topology, or a multi-hop network where each Sensor Node relies on its neighbours to forward its sensory data to the respective Sink Node.

Application Areas

                    Sensor networks have been adopted in a wide set of scenarios and applications where proper data management can be deemed of high importance. Some of the application areas where deployments of sensor networks with advanced capabilities are popular are the following:

• Health Monitoring: Biometric sensors are usually used for collecting and monitoring data regarding patients, administrating issues in hospitals, provision of patient care as well as for supporting the operation of special chemical and biological measurement devices (e.g. blood pressure monitoring). The collected sensory data are also stored for

historical reasons in order to be used for further survey on disease management and prognosis. In many cases, efficient representation and correlation of the acquired data

enable doctors and students to extract useful conclusions.

• Meteorology and Environment Observation: Environmental sensors are used for weather forecasting, wildfire and pollution detection as well as for agricultural purposes. Special observation stations collect and transmit major parameters which are used in the procedure of decision making. For example, in agriculture, air temperature, relative humidity, precipitation and leaf wetness data are needed for applying disease prediction models, while soil moisture is crucial for proper irrigation decisions towards understanding the progress of water into the soil and the roots.

• Industrial applications: Different kind of sensors are deployed for serving industries including aerospace, construction, food processing, environmental & automotive. Applications are being developed for tracking of products and vehicles in transportation companies, satellite imaging, traffic management, monitoring the health of large structures such as office buildings and several other industry-specific fields.

• Smart Homes: Home automation applications are being developed in order to support intelligent artifacts and make the users’ life more comfortable. Special sensors are attached to home appliances while the created sensor network can be managed or monitored by remote servers accessible via the Internet (e.g. user’s office, police office and hospital etc). Sensor networks also play a significant role on facilitating assisted living for the elderly or persons in need of special care.

• Defense (Military, Homeland Security): Sensors are also used for military purposes in order to detect and gain as much information as possible about enemy movements, explosions, and other phenomena of interest. Battlefield surveillance, reconnaissance (or scouting) of opposing forces, battle damage assessment and targeting are some of the fields where large sensor networks have been already deployed.

Sensor Web: Data and Services in a Sensor Network

                                   The term Sensor Web is used by the Open Geospatial Consortium (OGC) for the description of a system that is comprised of diverse, location aware sensing devices that report data through the Web. In a Sensor Web, entire networks can be seen as single interconnected nodes that communicate via the Internet and can be controlled and accessed through a web interface. Sensor Web focuses on the sharing of information among nodes, their proper interpretation and their cooperation as a whole, in order to sense and respond to changes of their environment and extract knowledge. Hence, one could say that the process of managing the available data is not just a secondary process simply enhancing the functionality of a

Sensor Web, but rather the reason of existence of the latter. Data storage can be either external (all data are collected on a central infrastructure), local (every node stores its data locally) or data-centric (a certain category of data is stored to a predefined node). External storage is not considered as a viable solution, because of the high energy cost of data transmission from each sensor node to the central infrastructure. Local storage overcomes this drawback, since every node stores only self generated data. The option of local storage is also referred to as Data-Centric Routing (DCR), where a routing algorithm is needed in order to answer a query or to perform an aggregation, focusing on minimizing the cost of communication between sensor nodes.

Knowledge Management in Sensor Networks

                              As stated earlier, the rapid development and deployment of sensor technology involves many different types of sensors, both remote and in-situ, with diverse capabilities. The absence of ontological infrastructures for high-level rules and queries restricts the potential of end users to exploit the acquired information, to match events from different sources and to deploy smart applications which will be capable of following semantic-oriented rules. Current efforts at the OGC Sensor Web Enablement (SWE) aim at providing interoperability at the service interface and message encoding levels. Sensor Web Enablement presents many opportunities for adding a real-time sensor dimension to the Internet and the Web. It is focused on developing standards to enable the discovery, exchange, and processing of sensor observations. The functionality that OGC aims to supply a Sensor Web with, includes discovery of sensor systems, determination of a sensor’s capabilities and quality of measurements, access to sensor parameters that automatically allow software to process and geo-locate observations, retrieval of real-time or time-series observations, tasking of sensors to acquire observations of interest and subscription to and publishing of alerts to be issued by sensors or sensor services based upon certain criteria.

                                   Technologies and standards issued by the World Wide Web Consortium (W3C) will be used in this context to implement the Semantic Sensor Web (SSW) vision, an extension of the Sensor Web, where sensor nodes will be able to discover their respective capabilities and exchange and process data automatically without human intervention. Components playing a key role in Semantic Sensor Web are ontologies, semantic annotation, query languages and rule languages. Rules can be defined using SWRL (Semantic Web Rule Language) and additional knowledge can be extracted by applying rule-based reasoning. Moreover, complex queries written in SPARQL Query Language for RDF - a W3C recommendation (or equivalently, a standard for the Web) - can be submitted to the Sensor Web for meaningful knowledge extraction and not just for simple retrieval of sensor readings. The application of these technologies will transform the Sensor Web Enablement service standards to Semantic Web Service interfaces, enabling sensor nodes to act as autonomous agents being able to discover neighbouring nodes and communicate with each other.

Current Approaches         

                        Many approaches are available today for managing sensor networks, regarding Data Management in Sensor Networks using Semantic Web Technologies 107 especially the aggregation and processing of data and several architectures have been proposed that provide services to the end user through the exploitation of the collected data. Existing approaches combine data from sensors in order to carry out high-level tasks and offer to the end user a unified view of the underlying sensor network. They usually provide a software infrastructure that permits users to query globally distributed collections of high bitrate sensors’ data powerfully and efficiently. Following this approach, the SWAP framework proposes a three tier architecture comprising a sensor, knowledge and a decision layer, each one of them consisting of a number of agents. Special care is taken for the semantic description of the services available to the end user, allowing the composition of new applications.

CHALLANGES FOR THE DATABASE COMMUNITY

                              Given the view of the sensor network as a huge distributed database system where each sensor node corresponds to a database site that holds part of the data, we would like or adapt existing techniques from distributed and heterogeneous database systems for the sensor network environment. But at close investigation, we can distinguish four major differences between sensor networks and traditional distributed and heterogeneous database systems. Physical Characteristics. Sensor networks have physical characteristics that are very different from regular desktop computers or dedicated equipment in data centers. Sensors might fail at any time; the networking layer might only provide very weak quality of service, and the sensor nodes have strict resource limitations such as limited memory, computational and battery power. Query processing has to be aware of these physical constraints. One way of thinking about such constraints is the analogous interaction with the operating systems in traditional database systems.

                               Database systems bypass the operating system buffer to have direct control over the disk. For a sensor network database system, the analogous resource is the networking layer, and for intelligent resource management we have to ensure that the query processing layer is tightly integrated with the networking layer. We can distinguish several types of queries in a sensor network. Long-running queries deal with the status of the sensor network over a user-defined time period. Other queries are ad-hoc or snapshot queries that query the current status of the sensor network. Strategies for evaluating these two types of queries are likely to be very different: In long-running queries, we can pay up-front a higher cost that can be amortized over the lifetime of the query.                            

                             Due to inherent resource limitations of sensor networks, users should be able to trade off the accuracy of a query answer versus the quantity of resources used to compute the query answer. As a simple example, assume that the sensor network consists of N temperature sensor nodes. To accurately compute the average temperature, all sensor nodes need to be contacted and their temperatures aggregated. But the user might be sufficiently confident in the average of M << N sensor readings, given that the sensors chosen are a random sample of the overall set of sensors. Computing the average of M sensor readings requires much less energy since only a small subset of all sensors is contacted. This is analogous to the computation of an aggregate through a sample in a database. Further research is necessary to understand these tradeoffs in detail as it is not obvious how to select a truly random sample of sensor nodes that satisfy a given geographic constraint without complete knowledge about the sensor network at a central query optimization node. Data Streams. Sensors produce data continuously in data streams, and sensor nodes have only limited memory and computational resources. We need to develop new query processing techniques for the online processing of data streams that do not assume that relations are materialized on secondary storage. Important for data stream processing will be intelligent data reduction at individual sensor nodes through the computation of stream aggregates. In addition to the computation of such statistics, we need to be able to process these synopsis data structure themselves when we combine synopsis data from several sensors.

                             Many sensor networks will include actuators — devices that allow manipulation of properties of the physical world; simple examples are temperature controls, door locks, or light switches. Scalable, distributed trigger management is a considerable research challenges for large-scale monitoring and control sensor systems.

Data Layer

                            This layer handles raw sensor data discovery, collection and aggregation in a central entity. Efficient data aggregation is crucial for reducing communication cost, thereby extending the lifetime of sensor networks. Based on the topology of the network, the location of sources and the aggregation function, an optimal aggregation structure can be constructed. Optimal aggregation can be defined in terms of total energy consumption, bandwidth utilization and delay for transporting the collected information from simple nodes to the sink nodes. Data gathering can be realized following structured or structure-free approaches:

• Structured approaches are suited for data gathering applications where the sensor nodes are following a specific strategy for forwarding the data to the sink nodes. Due to the unchanging traffic pattern, structured aggregation techniques incur low maintenance overhead and are therefore suited for such applications. But, in case of dynamic environments, the overhead of construction and maintenance of the structure may outweigh the benefits of data aggregation. Furthermore, structured approaches are sensitive to the delay imposed from the intermediate nodes, the frequency of the data transmission and the size of the sensor network. The central entity is responsible for the discovery of new nodes and the specification of the data acquisition policy. The data acquisition can be event-based where data are sent from the source and a method is called to collect them (serial ports, wireless cameras) or polling based where the central node periodically queries the data from the managed sensors.

 Processing Layer

                        Due to the raw nature of sensory data and the fact that it cannot provide us with high-level information extraction, several XML-based models are being used in order to interpret it. This will leverage its usability, allow further processing and finally make it meaningful for the end user. Proper processing is necessary, especially in cases of aggregation of data from many heterogeneous sources and the need for discovery of possible correlations among the aggregated data. Furthermore, the processed data can be distributed to other network devices (e.g. PDAs) without the need for sensor-specific software. Different XML templates can interpret in a different way the sensory data according to the application related view. The aggregated data has to be processed and integrated in a manner that shortens the data    

exchanging transactions. Integrating the data and transforming it into an XML (possibly a Sensor) format makes it meaningful for the end user. Initially, the Processing Layer integrates the bulk of the incoming data.

                               It is not necessary neither optimal, in certain cases, to maintain the total amount of data. Consider for instance a sensor network consisting of some dozens of sensors measuring the temperature over a field. While keeping track of the temperature levels is useful, processing every single datum originating from every single sensor is not needed. Such practice would overload the network, augment its maintenance needs and consequently decrease its autonomicity. Moreover, the volume of the archived information would soon require a substantial storage capacity. Aggregated reports (such as the maximum or an average of the values reported) may be sufficient to describe the conditions that are present in the area of interest. Subsequently, the integrated information collected by the sensors has to be forwarded to the upper Semantic Layer. In order for this to be achieved, the information needs to be encapsulated in messages suitable for further machine processing.

Semantic Layer

                  The Semantic Layer abstracts the processed outputs from the heterogeneous, low-level data sources such as sensors and feature extraction algorithms, combined with metadata, thus enabling context capturing in varying conditions. Context annotation is configured through application-specific ontologies and it can be automatically initiated without any further human intervention. It must be noted that the Semantic Layer is not an indispensable part of sensor network architecture, in the same way that semantics do not need necessarily to be part of systems.

CONCLUSIONS

                        This paper outlines a research program that addresses fundamental problems in sensor networks:

                         Data streams, uncertainty about sensor measurements, query processing, and trigger management. While developing techniques that address the three problems above, we must not forget that scalability of the techniques with the size of the network, the data volume, and the query workload is an intrinsic consideration to any design decision. I believe that sensor networks are a research area with challenging data management problems for years to come.

REFERENCES

[1] M. Balazinska, A. Deshpande, M. J. Franklin, P. B. Gibbons, J. Gray, M. Hansen, M. Liebhold, S. Nath, A. Szalay, V. Tao, Data Management in the Worldwide Sensor Web, IEEE Pervasive Computing, p. 30-40 (2007).

[2] K.W. Fan, S. Liu, P. Sinha, Structure-Free Data Aggregation in Sensor Networks, IEEE Transactions on Mobile Computing, p.929-942 (2007). 116 A. Zafeiropoulos, D.E. Spanos, S. Arkoulis et al.

[3] V. Cantoni, L. Lombardi, P. Lombardi, Challenges for Data Mining in Distributed Sensor Networks, 18th International Conference on Pattern Recognition (ICPR'06), p. 1000-1007 (2006).

[4] P. Sridhar, A.M. Madni, M. Jamshidi, Hierarchical Data Aggregation in Spatially Correlated Distributed Sensor Networks, World Automation Congress (WAC '06), p.1-6

(2006).

[5] K. Romer, F. Mattern, The Design Space of Wireless Sensor Networks, IEEE Wireless Communications, p. 54-61 (2004).

[6] S. Rajeev, A. Ananda, C. M. Choon, O. W. Tsang. Mobile, Wireless, and Sensor Networks - Technology, Applications, and Future Directions, John Wiley and Sons,

2006.

[7] C. Reed, M. Botts, J. Davidson, G. Percivall, OGC® Sensor Web Enablement: Overview and High Level Architecture, IEEE Autotestcon, 2007, p.372-380 (2007).