Doing Science in the Cloud
Dario Ogrizović, Zlatan Car, Gordan Janeš, Božidar Kovačić · 2013
Acceleration of scientific discovery in Cloud Computing environment presents a viable opportunity for science and engineering with benefits compared to traditional High Performance Computing, especially for scientific problems that can be run with a high degree of parallelism and at low scale. However, communication intensive scientific problems which require low latency and high bandwidth interconnections and parallel file systems doesn’t perform well in the Cloud. The largest IaaS cloud providers still use commodity networking and storage devices which cause significant performance variability and therefore they are unable to provide a completely satisfactory environment for some scientific HPC applications. There are also smaller dedicated HPC Cloud providers with optimized virtualization models, high bandwidth, and low latency interconnections which improve scientific HPC application performance.