This paper introduces a model for sensor databases and describes the design and implementation of the COUGAR sensor database system. Sensor networks are increasingly used for monitoring and surveillance applications, where users issue long-running queries over stored data and sensor data. Existing systems are centralized and lack flexibility, as they extract data in a predefined way and do not scale well. In contrast, the proposed sensor database system allows queries to dictate which data is extracted from sensors. The model represents stored data as relations and sensor data as time series. Each long-running query defines a persistent view that is maintained during a given time interval. The COUGAR system extends the Cornell PREDATOR object-relational database system, modeling each sensor type as an Abstract Data Type (ADT). Signal-processing functions are modeled as ADT functions that return sensor data. Long-running queries are formulated in SQL with minimal modifications. The system includes a new mechanism for executing sensor ADT functions to support long-running queries. The paper discusses the challenges of processing sensor queries, including the need to handle asynchronous and high-latency signal-processing functions, and the difficulty of expressing queries involving time-based aggregates and correlations. To address these issues, the paper introduces virtual relations, which are tabular representations of sensor ADT functions. Virtual joins are used to execute ADT functions that do not return results in a timely manner. These virtual relations are append-only and naturally partitioned across devices, enabling distributed query processing. The COUGAR system is implemented on top of the WINS infrastructure and is designed to handle large-scale sensor networks. The system supports distributed query execution and allows sensor database systems to leverage computing resources on sensor nodes. The paper also discusses related work, including other sensor database systems and wireless sensor network projects. The authors conclude that sensor database systems are a promising area for database research, with the potential to provide flexible and scalable access to large collections of sensors. The COUGAR system represents a first step towards such a system, with future work focusing on improving query processing and handling sensor and communication failures.This paper introduces a model for sensor databases and describes the design and implementation of the COUGAR sensor database system. Sensor networks are increasingly used for monitoring and surveillance applications, where users issue long-running queries over stored data and sensor data. Existing systems are centralized and lack flexibility, as they extract data in a predefined way and do not scale well. In contrast, the proposed sensor database system allows queries to dictate which data is extracted from sensors. The model represents stored data as relations and sensor data as time series. Each long-running query defines a persistent view that is maintained during a given time interval. The COUGAR system extends the Cornell PREDATOR object-relational database system, modeling each sensor type as an Abstract Data Type (ADT). Signal-processing functions are modeled as ADT functions that return sensor data. Long-running queries are formulated in SQL with minimal modifications. The system includes a new mechanism for executing sensor ADT functions to support long-running queries. The paper discusses the challenges of processing sensor queries, including the need to handle asynchronous and high-latency signal-processing functions, and the difficulty of expressing queries involving time-based aggregates and correlations. To address these issues, the paper introduces virtual relations, which are tabular representations of sensor ADT functions. Virtual joins are used to execute ADT functions that do not return results in a timely manner. These virtual relations are append-only and naturally partitioned across devices, enabling distributed query processing. The COUGAR system is implemented on top of the WINS infrastructure and is designed to handle large-scale sensor networks. The system supports distributed query execution and allows sensor database systems to leverage computing resources on sensor nodes. The paper also discusses related work, including other sensor database systems and wireless sensor network projects. The authors conclude that sensor database systems are a promising area for database research, with the potential to provide flexible and scalable access to large collections of sensors. The COUGAR system represents a first step towards such a system, with future work focusing on improving query processing and handling sensor and communication failures.