Service-Oriented Software Architecture for Sensor Networks Jan Blumenthal University of Rostock 4. IuK Tage Rostock Rostock, 20 th June 2003

2 Outline Introduction to sensor networks Requirements Software Engineering Approach Conclusion

3 Evolution of sensor nodes Smart Dust (UCB) Piece of Silicon Sensor, Actuator Battery Processor HF Picoradio (UCB) WINS (UCLA) PC Blue-Node (UoR)

4 Characteristics of Sensor Nodes Limited capacity of –Battery (Lifetime: days – 10 years) –Processing capabilities –Transmission range (5…20 meters) Data rates: Bit/s … KB/s Transmission methods: –Bluetooth –ZigBee –nanoNET One specific application Price: some cents Coming up sensor nodes

5 Characteristics of Sensor Networks Properties: Wireless sensor nodes connect together autonomously Distributed organization Contain one or more data sinks Huge number of nodes to compensate transmission range (density: nodes/m 2 ) Nodes may move around Tasks: Cooperative data acquisition: –Cooperative data collection and preprocessing –Running distributed applications regionally –Forwarding of results to higher application levels Adaptive reactions on environment modifications Network has to behave robustly and fault-tolerant to: –Failure on single nodes –Transmission errors and distortion through obstacles –Intrusion and jamming

6 Applications Stress Monitoring Measurements in Fluids Intra Corporal Measurements Sensor Node Protecting flood dam with sandbags

7 Requirements of Sensor Networks 1.Self-Organization –Ad hoc network formation –Autonomous connection establishment 2.Network Maintenance –Addressing of nodes –Routing facilities –Compensation of network failures 3.Cooperative Processing –Context awareness –Location positioning and location awareness –Preprocessing of collected raw data 4.Security Mechanisms 5.Energy Optimizations

8 Context Awareness Infrastructure context –Refers to the infrastructure around the node –Example: Perception of bandwidth and reliability Domain context –Refers to the current environment of the node –Example: Knowledge of next data sink in the network Data sink node Simple node Transmission range Context Awareness: Adapts the behavior of the node to the current environment

9 Transfer to Measurement the field strengths of and Triangulation / multilateration Transfer of determined positions to data sink Max. 6 transmissions Decentralized positioning Cooperative Algorithms Reduction of data by preprocessing and aggregation Minimization of data stream to data sinks Example: Positioning

10 Network Maintenance Attribute based addressing –Assignment of data to location of data mining –Location awareness necessary –ID based addressing unfavorable Random node distribution Routes become obsolete quickly due to mobility –Example: Temperature at location (x, y) ? Cooperative network behavior –Data aggregation –Address resolution / location awareness –Adaptive routing strategy depends on mobility of nodes Communication reduction / avoidance –Connectionless transmission –Prevent network traffic through data aggregation –Event based communication in contrast to request / reply –Polling avoidance

11 Services Discovering and using services Cascading of services without previous knowledge of each other Definition: Little program accessed by standardized interfaces over the network. Service infrastructure for remote access: Service Proxy Physical LAN Bluetooth Surrogate Host Proprietary protocol Sensor A Sensor B Sensor C Surface profile service Surface profile ? Client Request TCP/IP Reply

12 Software Engineering Hides the complexity of the distributed system Standardized API to node application Provides access to services on remote nodes Network Maintenance Services Hardware Abstraction Cooperative Algorithms Context Awareness Middleware Common Middleware : Goal: Same middleware on different platforms

13 Middleware Properties Scalable Middleware Customization of data types during compile time Removing of unused components Generic Middleware Platform independent middleware leads to increased number of complex interfaces Adaption of interfaces instead of programs void setBaudrate(int handle, int baudrate) { hardware_addr=getIOAddress(handle); hardware_addr->BTR0=baudrate; } void setBaudrate(int baudrate) { // getIOAddress(handle); BTR0=baudrate; } Non Generic Middleware Generic Middleware Savings: parameter delivery, function call, stack operation, return value assignment, field operation Valid only, if one interface present

14 Middleware Properties II Adaptive Middleware Mobility and changes in infrastructure require adaptions to the software –Cluster head selection requires additional routing software –Changing measurement methods leads to new programs Exchange and run components dynamically –Start of new services Reflective Middleware Ability to understand and influence itself Application layers are not affected Inspection:Analyze behavior via logging / debugging Adaptation:Changing behavior of internal layers. Example:Customizing routing strategy depending on mobility To overcome all the mentioned requirements we designed a software model !!

15 Definitions Measurements and storage of data Hardware dependant Example: –Software driver for temperature sensor Sensor Application Application specific parts Middleware (routing, discovering nodes) Contains sensor application Node Application Sensor Network Application Describes main tasks of the entire network Contains several node applications Optional administration interface

16 Node Application Internal interface (generic middleware) No individual high level application, because: Nodes provide services only to the network Focus: Success of sensor network application instead of node application Cooperative behavior of nodes Sensor application Dynamic components

17 Example Sensor Node Model ARM Microcontroller Temperature Sensor 68HC11 Microcontroller Pressure Sensor 68HC11 µC No Sensor

18 Sensor Network Application systemwide interfaces Middleware and Operating System are compiled and optimized to each hardware Sensor Network Application –Composition of all node services Optional administration terminal to manage the entire network different node hardware boot, logging customized node‘s software

19 Sensor Network Design Compile/Link Design & Edit Distribute Execute/Administrate Development Studio Resource Hardware driven Monitoring Evaluation Optimization Components Changes to common design methods. Create hardware-optimized software components (driver, operating system) Create hardware-independent software components (middleware, services) Combining of predefined components Source code generation Removing unused components Optimization of interfaces Optimization to node‘s hardware Distribution of nodes in different environments Monitoring the execution Creation of logfiles Evaluation of logfiles

20 Conclusion New challenges of software development in sensor networks Optimization of node‘s software based on –Interface design –Adaptive and reflective middleware Design of service-oriented middleware in sensor networks Future work –Realization of presented software concept –Development studio

Thank you Any questions ?

22 Our Approach Adaption of program to the application task instead adaption of application task to the program Adaption of interfaces to the program instead adaption of the program to the interfaces Application-dependent customized software components instead multipurpose software on each node

23 Software Architecture Desktop Model Resource optimized Application Specific OS Modules Middleware Application Middleware Application Operating System Hardware Sensor Node Independent applications Standardized middleware Standardized operating system Example: Jini Independent applications Standardized middleware Standardized interfaces Unused components removed No independent applications Customized middleware Proprietary interfaces Optimized driver software Not optimized (Partly) optimized modules