Recently, a new class of data-intensive applications in which data are continuously and dynamically updated is emerging. However, traditional database management systems (DBMS) are designed to maximize system throughput rather than support for continuous data updates and provide short and predictable response times for transactions. In addition to continuous updates, there is another distinct class of transactions in these applications called continuous queries, which are different from traditional transactions. This project explores a number of new research issues involved in continuous queries with updates and proposes a new transaction processing protocol in main-memory sensor databases. In particular, it proposes algorithms to improve transaction response time by reducing overheads in transaction management, to improve timeliness by exploiting application semantics, and to improve data availability to continuous queries by caching a temporal consistent view of the data streams. To understand the behaviour and the performance of the protocol, a series of simulation experiments will be conducted to evaluate the algorithms.