s from operational concerns. Building CNAs is the aim of many new software projects, but also of migration projects where an existing system should be transformed to a CNA. A key aspect of such a migration is the architectural transformation to the fine-grained architecture of a CNA which typically means decomposing the existing system.
However, such a decomposition is dificult and rarely supported by structured
approaches or tools. Therefore, it would be beneficial to explore possible structured
approaches for the transformation of existing software systems to the fine-grained
cloud architecture of CNAs to support software engineers engaged in such projects.
This support will remain valuable in the future, because even today it argued
that it is a reasonable procedure to start of with a monolithic architecture [
A promising field of research in this regard is Model-Driven Engineering (MDE)
[
To explore the applicability of MDE for the migration of existing software towards fine-grained cloud architectures, the following research questions are investigated: – Which model or collection of models can be used to represent an existing software architecture and a fine-grained cloud architecture? – What are rules and patterns for the transformation to a fine-grained cloud architecture and how can they be applied to the identified model or collection of models? – How can the generation of models from code, the model-to-model transformation, and the generation of code from models be efectively supported by tools? 2
Various research directions can be identified covering diferent aspects of the
proposed research questions. Regarding modeling languages for the cloud, the
available languages [
In order to answer the given research questions, this position paper proposes a model-driven migration of existing software systems to fine-grained cloud architectures. The basic idea is that a model can be generated representing the existing system. This model can then be (semi-)automatically transformed to a model of a fine-grained cloud architecture. Finally, artifacts can be generated from the transformed model to facilitate the implementation of the new architecture. Because implementing this approach as a whole is ambitious, it should be separated into diferent concerns, as represented by the corresponding research questions. First, a modeling language or collection of languages needs to be defined. It should enable the modeling of an existing application, the identification of possibilities for deriving fine-grained components, as well as representing the resulting fine-grained cloud architecture. Therefore, a key challenge for the modeling language is to find the right level of abstraction. Next, rules and patterns for splitting an application into more fine-grained components should be defined and represented in model transformation rules. Finally, because tools are seen as a key part of MDE, tools should be provided for creating a model from code, transforming the model and generating artifacts for the new architecture.
The approach focuses on the actual execution of a migration. However, it should be noted that before the actual execution, a careful analysis of the necessity and usefulness of a migration should be conducted. Not every existing application should be transformed to a CNA and evaluating the applicability of a transformation is another important research question. Sometimes also hybrid applications are sensible where only parts of the existing system are migrated to be cloud-native. The approach could provide additional insights for this question as well, but to keep the approach focused, it is assumed that a prior evaluation has been done.
That said, there are several further challenges which need to be taken into account. There exists a large technological heterogeneity for existing applications and especially cloud-native applications. The aim for this approach would be to make the models for the existing system and the fine-grained cloud architecture as generic as possible. Technological diferences are then addressed inside the steps for generating a model from code and generating artifacts, e.g., through profiles. Furthermore, when thinking about possibilities to split an application into smaller components, a purely technical view might suggest several diferent possibilities with potentially conflicting outcomes. In such cases the business view rather than the technical view is crucial to make reasonable decisions. The approach could therefore be semi-automatic and suggest diferent options during the model transformation phase which could then be decided upon using business knowledge.
As a starting point, an exemplary application is manually transformed. During the manual transformation, requirements for the modeling language and first transformation rules can be identified. For this exemplary application a specific technological stack needs to be chosen. The diferences with regard to other technological possibilities, are considered afterwards. Further applications or successful migration projects are then considered to iteratively evaluate and improve the approach.
In conclusion, transforming applications into fine-grained CNAs is an ongoing challenge for developers. Providing support for this task through structured approaches and tooling would therefore be beneficial to developers and to research on CNAs as a whole.