evolution@home: Experiences with Work Units That Span More than 7 Orders of Magnitude in Computational Complexity

Individual-based models in evolutionary biology easily lead to multi-parameter applications that need global computing power to exploit their full potential. Mainly due to varying population size parameters, they easily generate computational complexities from less than a second to more than 100 years in case of the Simulator005 of evolution@home. The poorly understood biology of the system leads to automated predictions that may be way off. This report describes first experiences of a global computing system, where users can choose between tasks of different complexity. Besides theoretical complexity limits of tasks that fit global computing, choices of users are analyzed. Potential of incomplete results to increase prediction accuracy is discussed as well as benchmarking computer systems that vary nearly 2 orders of magnitude in their idle processing power. Finally, prediction accuracy is analyzed with the help of a newly defined parameter: error of magnitude. It is concluded, that global computing has great potential for projects with poorly predictable single-run-complexities, if frameworks are designed to allow users to choose their commitment, and if they make use of incomplete results to improve predictions.