Performance prediction of configurable software

What is it about?

When configuring hardware and software products, it is helpful to know their performance (e.g. cost, fuel usage, image processing throughput, or flash/memory usage) during configuration. This way, users immediately see how configuration options affect the product and can make informed decisions when considering whether to include optional features or when choosing between alternatives. We discuss whether the models used to predict product performance should be part of the configuration option definitions or kept separately, and argue that separate performance prediction models should be preferred.

Featured Image

Photo by Malte Helmhold on Unsplash

Photo by Malte Helmhold on Unsplash

Why is it important?

After decades of working with configurable hardware and software products, there is still no consensus on which variability modeling language should be used to define configuration options and dependencies between them. This also applies to language features that are used for performance prediction. We show that variability modeling languages do not need to consider performance prediction, as separate performance models are often more accurate and flexible than those integrated within a variability model. Thus, there is one less aspect to worry about when discussing the design of variability modeling languages.