Friday, August 5, 2011

Research Targets Computational Experiments in the Cloud

HPC in the Cloud (07/27/11) Nicole Hemsoth

A St. Andrews University research team recently launched an investigation into the viability of running computational experiments on virtualized hardware. Researcher Lars Kothhoff says in an interview that reliability and reproducibility of results is critical to such research, especially when it involves deploying complex systems that solve artificial intelligence problems. "Our results show that there is no inherent disadvantage to using virtualized resources for such experiments as long as the results are carefully evaluated and statistical methods used to identify outliers and establish confidence in the conclusions drawn from the results," he notes. Kothhoff says the best case for running central processing unit (CPU)-intensive tasks on cloud resources is likely one that involves many experiments or experiments that can be easily distributed across multiple virtual machines when there are not sufficiently available physical resources. He says that getting reliable CPU timings is a substantial difficulty only for the small types of virtual machines in a cloud. "For larger types, the reliability becomes as good as on physical hardware or even better," Kothhoff says.