Evolving Multi-Tenant SaaS Cloud Applications Using Model-Driven Engineering

Assylbek Jumagaliyev, Jon Whittle, Yehia ElkhatibME@MoDELS 2016

Introduction

In the Software-as-a-Service (SaaS) paradigm, multi-tenancy is the cornerstone of operational efficiency, allowing multiple customers (tenants) to share underlying hardware and software resources. However, this shared nature introduces significant complexity when applications must evolve. Developers face a dilemma: how to implement custom workflows or specialized security architectures for one tenant without destabilizing the system for others. As applications scale, the interplay between variability (different configurations) and evolution (changing requirements) creates a management burden that traditional manual coding cannot easily sustain. This paper proposes a hybrid framework combining Software Product Lines (SPL) and Model-Driven Engineering (MDE) to address these challenges.

A Hybrid Framework for Evolution

The proposed approach treats architectural design decisions as "features" within a product line, allowing for systematic customization and updates. The workflow integrates SPL's ability to manage variability with MDE's capacity for automated system generation.

The proposed MDE-based framework for SaaS evolution.

The process begins with a Feature Model that captures both core and variant requirements. Core features are modeled using standard UML, while variant features are modeled independently using MATA (Modeling Aspects using a Transformation Approach). MATA serves as a composition engine, merging variant models into the core UML model while automatically detecting structural conflicts or dependency violations. This separation of concerns ensures that tenant-specific customizations remain modular and manageable.

Layered Feature Modeling

To organize the vast array of requirements in a cloud environment, the authors utilize a three-layered feature hierarchy. This structure provides a clear roadmap for evolution by categorizing features based on their abstraction level:

  1. Capability Layer: Defines high-level functional requirements (e.g., "Create Surveys").
  2. Domain Technology Layer: Specifies the architectural choices for implementation (e.g., "Data Storage" strategies).
  3. Implementation Technique Layer: Details the specific technologies employed (e.g., "SQL Database" vs. "Blob Storage").
The three-layered feature model for the Surveys case study.

By mapping high-level capabilities to specific implementation techniques, developers can evolve a tenant's setup—for instance, moving from a shared database to an isolated one—simply by re-selecting features within the model.

Case Study: The "Surveys" Service

The framework was validated using a "Surveys" service case study, a multi-tenant application where tenants create and analyze online surveys. The authors demonstrated two critical evolution scenarios:

  • Scaling: Transitioning from a single shared database to multiple instances to accommodate a growing tenant base.
  • Feature Enhancement: Upgrading a tenant from basic Blob storage to a complex SQL database to enable advanced data analysis tools.

The results showed that by re-selecting models from the variability pool, valid application architectures could be generated efficiently, proving the feasibility of model-driven evolution in SaaS.

Conclusion and Future Work

The integration of SPL and MDE provides a robust foundation for managing the lifecycle of multi-tenant applications. While the current framework successfully generates composed UML models and detects conflicts, full automation of source code generation remains a primary goal for future development. The authors plan to refine the MATA tooling to be more "SaaS-aware" and conduct broader comparative studies to further quantify the productivity gains offered by this model-driven approach.

Summary updated

BibTeX
@inproceedings{jumagaliyev2016evolving,
  author = {Assylbek Jumagaliyev and Jon Whittle and Yehia Elkhatib},
  title = {{Evolving Multi-Tenant SaaS Cloud Applications Using Model-Driven Engineering}},
  booktitle = {ME@MoDELS},
  year = {2016},
}
Figure