AUTHOR(S):Tobias Lindinger, Nils gentschen Felde, Dieter Kranzlmüller
ABSTRACT: This poster provides an analysis of the suitability of clouds for high-performance computing, thereby providing a qualitative evaluation of large-scale, virtualized infrastructures in contrast to specialized clusters or supercomputers. Typical problems arising in such virtualized HPC-scenarios are the unknown effectiveness of hypervisor implementations, as well as side effects of virtual machines influencing the performance of co-located virtual machines on the same physical host. This results in performance degradation due to either implementation overhead or using the same physical hardware components concurrently. We measured these effects qualitatively and quantitatively and assembled a general list of advantages and disadvantages of using virtualization approaches in HPC environments. The result is a set of recommendations on scenarios, which can be virtualized and which not for HPC purposes. The poster describes the setup of the tested scenarios and concrete results on the obtained performance, as well as the resulting recommendations.