Resreach On Model Transformation Method Based On MDA: From Class Diagram Model To Cassandra Date Model

With the coming of big data era, data are generated in an unprecedented speed. There is immeasurable value hidden in big data but which cannot be handled by traditional database techniques because they lack well horizontal scalability and then NoSQL databases emerged as times require. NoSQL stores offer a better capacity of squeezing value from big data. However, designing a proper schema for NoSQL databases is totally different from designing for relational databases and it would be such a blessing if this process can be automated.Here we use MDA to achieve this goal. MDA sees the process of software development as a series of model transformations. In this thesis we transformed class diagrams, a kind of UML diagrams that is widely used in object-oriented software development, into Cassandra data model. So far Cassandra is the most popular column-oriented database. The model transformation proposed in the thesis is based on meta-modeling mothed and can be separated into two phases. Phase one is about model-to-model transformation. In this phase class diagram instances are transformed into database schemas and both of them are represented in xmi files. Phase two is model-to-text transformation. Executive code will be generated at this phase.In order to achieve phase one transformation, the meta-models of the source model and the target model should be obtained at first. Some of existing work built meta-models for class diagrams but which did not in accordance with the definition of UML standard, which will lead to compatible problems. Nevertheless, the definition of UML standard is too complicated and contains lots of elements that are unnecessary for database schema generating. The meta-model for class diagrams is obtained through simplifying this definition and retains full compatibility with UML standard at the same time. For the sake of better performance of the generated schema, the inner storage structure of Cassandra is studied thoroughly. The construction of target meta-model and a part of design strategies of model mapping are also derived from this exploration.Query patterns must be considered when designing schemas for Cassandra, but that are some information that class diagrams do not have. In order to solve this problem, we designed an annotation system which allow annotations are added into the source model to provide extra auxiliary information for model transferences. When designing model mapping rules, the features of the source model and the target model must be considered carefully. For example, how to deal with various inter-class relationships in class diagram, given a class diagram instance how to generate a proper schema for Cassandra, etc. When the phase one transformation is complete, the phase two becomes much easier. In the last part, an example about employment management system is deployed to verify the two- phase transformation. The result proves that the model transformation discussed in this thesis is feasible and have the capacity to generate proper schemas for Cassandra.