A modeling language for multi-tenant data architecture evolution in cloud applications

Multi-tenancy enables efficient resource utilization by sharing application resources across multiple customers (i.e., tenants). Hence, applications built using this pat- tern can be offered at a lower price and reduce maintenance effort as less application instances and supporting cloud resources must be maintained. These properties en- courage cloud application providers to adopt multi-tenancy to their existing applications, yet introducing this pattern requires significant changes in the application structure to address multi-tenancy requirements such as isolation of tenants, extensibility of the application, and scalability of the solution. In cloud applications, the data layer is often the prime candidate for multi-tenancy, and it usually comprises a combination of different cloud storage solutions such as blob storage, relational and non-relational databases. These storage types are conceptually and tangibly divergent, each requiring its own partitioning schemes to meet multi-tenancy requirements. Currently, multi-tenant data architectures are implemented using manual coding methods, at times following guidance and patterns offered by cloud providers. However, such manual implementation approach tends to be time consuming and error prone. Several modeling methods based on Model-Driven Engineer- ing (MDE) and Software Product Line Engineering (SPLE) have been proposed to capture multi-tenancy in cloud applications. These methods mainly generate cloud deployment configurations from an application model, though they do not automate implementation or evolution of applications. This thesis aims to facilitate development of multi-tenant cloud data architectures using model-driven engineering techniques. This is achieved by designing and implementing a novel modeling language, CadaML, that provides concepts and notations to model multi-tenant cloud data architectures in an abstract way. CadaML also provides a set of tools to validate the data architecture and automatically produce corresponding data access layer code. The thesis demonstrates the feasibility of the modeling language in a practical setting and adequacy of multi-tenancy implementation by the generated code on an industrial business process analyzing application. Moreover, the modeling language is empirically compared against manual implementation methods to inspect its effect on developer productivity, development effort, reliability of the application code, and usability of the language. These outcomes provide a strong argument that the CadaML modeling language effectively mitigates the high overhead of manual implementation of multi-tenant cloud data layers, significantly reducing the required development complexity and time