Using an Open-Source Grid Framework and Virtualization for Embarrassingly Parallel Computations
Abstract
In this talk, we present the overall architecture and ideas driving the Climate@Home project. Building on the success of ClimatePrediction.net and SETI@Home, Climate@Home enlists citizen volunteers and NASA’s cloud services to donate their spare compute cycles to run the GISS ModelE climate model in a massively parallel fashion. NASA’s cloud services and project data management support will come from NASA’s Infrastructure as a Service (IaaS), powered by Nebula(TM). By using these donated cycles, scientists are able to run the model with significantly more configurations to help better determine the model’s sensitivity to its input parameters. Our architecture differs from the widely used BOINC platform (as used by ClimatePrediction.net and SETI@Home) in its use of open source virtualization technology, VirtualBox. By providing a standardized and highly controlled environment, we can ensure bit-wise reproducibility of computational results across heterogeneous platforms, helping to maintain the scientific integrity of the model experiments. While the initial experiments focus on climate, the system is designed to be flexible enough to apply to any “embarrassingly parallel,” computationally intensive tasks via the Java Parallel Processing Framework (JPPF) open source grid framework. The system provides a simple, generic configuration interface for scientists to create experiments and submit them to the grid. It also provides a pluggable architecture, allowing different projects to provide their own visualization capabilities. Finally, metrics and management features provide users with the ability to control and track contributions to their selected projects.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2010
- Bibcode:
- 2010AGUFMIN23B1362F
- Keywords:
-
- 1920 INFORMATICS / Emerging informatics technologies;
- 1999 INFORMATICS / General or miscellaneous