Geir Horn, Univ. of Oslo
Dr. Geir Horn hold s Cand. Scient in cybernetics and a PhD in mathematical machine learning, both from the University of Oslo, and he is now Head of European ICT Projects with the The Faculty of Mathematics and Natural Sciences and Department of Informatics at the University of Oslo. Geir has previous been Research Director with SINTEF Electronics and Cybernetics, and has worked with basic research with SIMULA Research Laboratory. Dr. Horn has also been coordinator and principal investigator for 19 European collaborative projects in the fields of High Performance Computing (HPC) and Cloud computing. Geir is now leading the research group on Scalable Computing within SIRIUS (https://sirius-labs.no/), a Centre for Research driven Innovation at UiO, and coordinating the Horizon 2020 project MORPHEMIC (https://www.morphemic.cloud/). Dr. Horn's research interest are combinatorial and dynamic optimization in stochastic systems, with a special focus on collective intelligence and learning in large scale distributed multi-agent systems.
Session
Persistent applications will experience variable demands and workloads, and Cloud resources may mitigate performance problems by allowing the necessary resources to be temporarily rented as needed. Cloud deployment decisions must consider not only the best suited providers but also the possibility to use beneficially hardware accelerators like GPUs and FPGAs should these be useful for the application and improve its performance. Since additional application resources may be needed at any time in response to the changed workload, or as a function of the data processed by the application, automated application management is desirable. This talk will cover the versatile MORPHEMIC platform capable of managing and optimizing the application over its lifetime using forecasting to predict and anticipate changes in the execution environment and proactively provide the Cloud resources timely to the application. Additionally, MORPHEMIC can deploy polymorphic application components to the hardware resources best apt for the application tasks at hand. Examples from real world applications will be given to show the benefits of the approach.