Abstract

Natural disasters drastically impact the society, causing emotional disorders as well as serious accidents that can lead to death. These kinds of disasters cause serious damage in computer and communications systems, due to the complete or partial destruction of the infrastructure, causing software applications that actually run on those infrastructures to crash. Additionally, these software applications have to provide a stable service to a large number of users and support unpredictable peaks of workloads. In this work, we propose a methodology to predict the performance of software applications designed for emergency situations when a natural disaster strikes. The applications are deployed on a distributed platform formed of commodity hardware usually available from universities, using container technology and container orchestration. We also present a specification language to formalize the definition and interaction between the components, services and the computing resources used to deploy the applications. Our proposal allows to predict computing performance based on the modeling and simulation of the different components deployed on a distributed computing platform combined with machine learning techniques. We evaluate our proposal under different scenarios, and we compare the results obtained by our proposal and by actual implementations of two applications deployed in a distributed computing infrastructure. Results show that our proposal can predict the performance of the applications with an error between 2% and 7%.