Feature Models at Run Time: Feature Middleware for Multi-tenant SaaS applications

Software product line engineering (SPLE) techniques revolve around a central variability model which in many cases is a feature model that documents the logical capabilities of the system as features and the variability relationships between them. In more traditional SPLE, this feature model is a result of domain analysis and requirement elicitation, while more recently this approach has been extended to represent also design-time variability, for example to document different ways to realize the same functionality. In many approaches, the feature model has run-time relevance as well. For example, in earlier work, we have used SPLE techniques to develop customizable multi-tenant SaaS applications, i.e. SaaS applications of which a single run-time instance is offered to many customer organizations (tenants), often with widely different requirements. In such systems, tenant customization is accomplished entirely at run time. In this paper, we present and explore the idea of promoting the feature model as a run-time artifact in the context of customizable multi-tenant SaaS applications, and we discuss the potential benefits in terms of the deployment, operation, maintenance, and evolution of these systems. In addition, we discuss the requirements this will impose on the development methods, the variability modeling languages, and the middleware.