Cache Sharing in QImPrESS Performance Models

The aim of the Q-ImPrESS project [Q-ImPrESS, 2009] is to develop methods and tools for predicting impact of changes in existing software systems on quality attributes such as performance, reliability and maintainability. The methods are centered around creating a model of the software system and performing the changes first in the model, transforming the model to a prediction model and solving the latter to obtain the desired quality estimates. Our goal is to extend performance prediction models used within Q-ImPrESS to include effects of implicit resource sharing, which are mostly neglected, yet often have significant impact on performance. This paper summarizes the current state of our work regarding modeling of processor cache sharing, highlights the effects we observed in past experiments, and outlines the upcoming work of incorporating some of these effects in suitable Q-ImPrESS performance models.