Towards the performance analysis of elastic systems with e-motions

Abstract

We use graph transformation to define an adaptive component model, what allows us to carry on predictive analysis of dynamic architectures through simulations. Specifically, we build on an e-Motions definition of the Palladio component model, and then specify adaptation mechanisms as generic adaptation rules. We show how the simulation-based analysis available in such a static definition can be extended in order to use the collected information on metrics such as response time, throughput and resource usage to adapt to the workload of the system and the environmental conditions. We illustrate our approach with rules modeling the scale in and out of servers, fired in response to the violation of specified constraints on the usage of resources. We evaluate this scenario by analyzing its performance, and discuss on its consequences in practice.