Systematic Design and Analysis of Customized Data Management for Real-Time Database Systems

Modern real-time data-intensive systems generate large amounts of data that are processed using complex data-related computations such as data aggregation. In order to maintain logical data consistency and temporal correctness of the computations, one solution is to model the latter as transactions and manage them using a Real-Time Database Management System (RTDBMS). Ideally, depending on the particular system, the transactions are customized with the desired logical and temporal correctness properties, which should be enforced by the customized RTDBMS via appropriate transaction management mechanisms. However, developing such a data management solution with high assurance is not easy, partly due to inadequate support for systematic specification and analysis during the design. Firstly, designers do not have means to identify the characteristics of the computations, especially data aggregation, and to reason about their implications. Design flaws might not be discovered early enough, and thus they may propagate to the implementation. Secondly, meeting more properties simultaneously might not be possible, so trading-off the less critical ones for the critical one, for instance, temporal correctness, is sometimes required. Nevertheless, trade-off analysis of conflicting properties, such as transaction atomicity, isolation and temporal correctness, is mainly performed ad-hoc, which increases the risk of unpredictable behavior.In this thesis, we address the above problems by showing how to systematically design and provide assurance of transaction-based data management with data aggregation support, customized for real-time systems. We propose a design process as our methodology for the systematic design and analysis of the trade-offs between desired properties, which is facilitated by a series of modeling and analysis techniques. Our design process consists of three major steps as follows: (i) Specifying the data-related computations, as well as the logical data consistency and temporal correctness properties, from system requirements, (ii) Selecting the appropriate transaction models to model the computations, and deciding the corresponding transaction management mechanisms that can guarantee the properties, via formal analysis, and, (iii) Generating the customized RTDBMS with the proved transaction management mechanisms, via configuration or implementation. In order to support the first step of our process, we propose a taxonomy of data aggregation processes for identifying their common and variable characteristics, based on which their inter-dependencies can be captured, and the consequent design implications can be reasoned about. Tool support is provided to check the consistency of the data aggregation design specifications. To specify transaction atomicity, isolation and temporal correctness, as well as the transaction management mechanisms, we also propose a Unified Modeling Language (UML) profile with explicit support for these elements. The second step of our process relies on the systematic analysis of trade-offs between transaction atomicity, isolation and temporal correctness. To achieve this, we propose two formal frameworks for modeling transactions with abort recovery, concurrency control, and scheduling. The first framework UPPCART utilizes timed automata as the underlying formalism, based on which the desired properties can be verified by model checking. The second framework UPPCART-SMC models the system as stochastic timed automata, which allows for probabilistic analysis of the properties for large complex RTDBMS using statistical model checking. The encoding of high-level UTRAN specifications into corresponding formal models is supported by tool automation, which we also propose in this thesis. The applicability and usefulness of our proposed techniques are validated via several industrial use cases focusing on real-time data management.