Engineering Model-Based Adaptive Software Systems

Adaptive software systems are able to cope with changes in the environment by self-adjusting their structure and behavior. Robustness refers to the ability of the systems to deal with uncertainty, i.e. perturbations (e.g., Denial of Service attacks) or not-modeled system dynamics (e.g., independent cloud applications hosted on the same physical machine) that can affect the quality of the adaptation. To build robust adaptive systems we need models that accurately describe the managed system and methods for how to react to different types of change.

In this thesis we introduce techniques that will help an engineer design adaptive systems for web applications. We describe methods to accurately model web applications deployed in cloud in such a way that it accounts for cloud variability and to keep the model synchronized with the actual system at runtime. Using the model, we present methods to optimize the deployed architecture at design- and run-time, uncover bottlenecks and the workloads that saturate them, maintain the service level objective by changing the quantity of available resources (for regular operating conditions or during a Denial of Service attack). We validate the proposed contributions on experiments performed on Amazon EC2 and simulators.

The types of applications that benefit the most from our contributions are web-based information systems deployed in cloud.