Machine Learning for Performance Prediction in Complex Software Systems - Detailseite

Complex software systems typically deliver their function through the interplay of multiple, conceptually independent components. This makes the prediction of the runtime behavior of such systems very difficult; core challenges are the increasing complexity of hardware, the interaction between workload and hardware, and resource contention caused by parallel workload. In this seminar, we specifically look at distributed workflow systems as often used for the large-scale analysis of scientific data sets. We focus on techniques from the field of machine learning to predict the behavior of single components and of entire workflows.

Das Seminar wird im im Wesentlichen als Blockseminar am Ende des Semesters stattfinden.