Executable Modeling of Deployment Decisions for Resource-Aware Distributed Applications

Abstract

The increasing popularity of virtualized services and cloud computing, offering elastic and scalable computing resources challenges software engineering methods by asking a number of new questions: How can we integrate deployment-specific information in the overall design of software applications? How can we express and compare deployment decisions in the design phase, so that performance diagnosis can happen early in the software development cycle? How do we design scalable applications?

This thesis proposes a methodology for the modeling and analysis of object-oriented distributed applications that are able to adapt to changes in client traffic, so that they can keep desired levels of performance. As a starting point for developing our methodology, we take Core ABS, an abstract, behavioral specification language that targets the executable modeling of concurrent, distributed and object-oriented systems. We extend Core ABS with deployment-specific information such as deployment architectures, flexible application-specific schedulers, deadlines at the application level, resource management primitives, and user-defined resource costs.

The extensions are developed following conventions of formal methods, combining rigorous definitions and formal semantics with a user-friendly Java-like syntax and tool support. By exploiting these language extensions, our methodology makes it possible to compare at the modeling level how a software behaves under different deployment choices. Such comparisons allow a better understanding of the trade-offs from different deployment choices, consequently better design decisions for distributed applications can be made early in the software development life cycle.

By integrating resource management and deployment decisions in the design phase, our methodology has the potential to improve the software engineering process for virtualized and scalable software applications running in the cloud, both with respect to quality and with respect to development and deployment cost.