Multi-formalism Models for Performance Engineering

Abstract

Nowadays, the necessity to predict the performance of cloud and edge computing-based architectures has become paramount, in order to respond to the pressure of data growth and more aggressive level of service agreements. In this respect, the problem can be analyzed by creating a model of a given system and studying the performance indices values generated by the model’s simulation. This process requires considering a set of paradigms, carefully balancing the benefits and the disadvantages of each one. While queuing networks are particularly suited to modeling cloud and edge computing architectures, particular occurrences—such as autoscaling—require different techniques to be analyzed. This work presents a review of paradigms designed to model specific events in different scenarios, such as timeout with quorum-based join, approximate computing with finite capacity region, MapReduce with class switch, dynamic provisioning in hybrid clouds, and batching of requests in e-Health applications. The case studies are investigated by implementing models based on the above-mentioned paradigms and analyzed with discrete event simulation techniques.