Thinking parallel: Multi-cores, virtual elasticity, and the application programmer
Abstract:
CPUs can offer a large number of cores and dedicated accelerators. Large data centres offers large number of virtual machines in the Cloud. At the exception of some highly optimised eScience applications, these resources are used to support high throughput computing of data parallel applications. Mixing true parallelism with the cloud is inherently complicated, and the application programmer cannot think parallel. This talk will discuss the issue and present some possible ways forward, hopefully stimulating an interesting discussion on future research directions.
Biography:Geir Horn is Cand. Scient. in Cybernetics and Ph.D in Computer Science, both degrees from the University of Oslo. Geir is currently Head of European ICT projects at the Faculty of Mathematics and Natural Sciences at the University of Oslo. Geir has previously held positions as senior scientist and research director at SINTEF in Oslo, before spending 4 years in more basic research at the SIMULA Research Laboratory. He has been working with European research for 20 years and has been coordinating 16 European collaborative projects ranging from smaller support actions to large integrated projects. His current research interests are on how to handle complexity and service choreography for large-scale distributed applications through adaptation, autonomic decisions, self-awareness, and emergence. In particular, Geir's focus is on games of learning automata for combinatorial optimisation in stochastic systems.