Geophysical fluid dynamics education and research

Project Details

Project Lead
Glenn Flierl 
Project Manager
Glenn Flierl 
Institution
Massachusetts Institute of Technology, Earth, Atmospheric, and Planetary Sciences  
Discipline
Earth, Atmospheric, and Ocean Sciences, n.e.c. (304) 

Abstract

We are developing and deploying infrastructure for integration of state-of the art numerical models running on cloud-computing clusters into classroom settings. Our middleware will (i) support the deployment of applications created by leading researchers on cloud computing facilities and (ii) facilitate the development of classroom modules for educational purposes. We will field test and showcase our approach by using it to develop a virtual-fluid-laboratory which will allow students to carry out and visualize numerical experiments that are analogues of classic laboratory experiments. Although the virtual fluid laboratory is our testbed, the enabling technology we propose is generic and has the potential to support an essentially unlimited number of science, engineering and math educational computing activities. Using the approaches we propose it will become possible for students, equipped with noth- ing more than a low cost laptop or "netbook" computer, to routinely interact with advanced science and technology applications developed by leading national educators and researchers. In conjunction with our development of an interactive virtual textbook and educational structures like MITx/Edx, and with the use of FutureGrid for appropriate US courses and commercial clouds worldwide, we can provide a practical and sustainable method for large numbers of students to use complex models in their studies.

Intellectual Merit

This work will expose students at all levels to tools and techniques of modeling as well as the use of compute clusters. Therefore, not only will students gain access to more sophisticated models for specific educational topics, but they will also gain familiarity with more advanced approaches to computational systems. Today the performance of EC2 type services for parallel computation is limited, but the concepts and strategies for using these systems can still be appreciated. Performance seems likely to increase in the future -- indeed we were recently contacted by the developers of the open source parallel library OpenMPI with an inquiry about collaborating to improve performance of our current EC2 demonstrations.

Broader Impacts

We believe that this work fills an important gap in the STEM education strategy of the nation. Increasingly computational modeling is a key part of any technical endeavor. In the past many innovations have been driven by increasing horsepower of desktop systems. However, in recent years desktop CPU clockspeed increases have flattened and performance boosts today are increasingly reliant on making computers "wider" [33] i.e. introducing multi-thread and process parallelism. This means that desktop computing needs to be supplemented by more powerful compute clusters in many endeavors, in order for practitioners to remain competitive. The work we propose here will expose students at all levels to tools and techniques of modeling on compute clusters. Therefore, not only will students gain access to more sophisticated models for specific educational topics through our work, but students will also gain familiarity with more advanced approaches to computational systems. Today the performance of EC2 type services for parallel computation is limited, but the concepts and strategies for using these systems can still be appreciated. Performance seems likely to increase in the future -- indeed we were recently contacted by the developers of the open source parallel library OpenMPI with an inquiry about collaborating to improve performance of our current EC2 demonstrations. As the NSF and other US agencies and commercial entities invest more and more in high-end computing and national network infrastructure, it is vital that we create an adequate pool of students who will be able to take maximal advantage of those resources when they join the workforce. This work will go a long way toward creating a foundation on which to build the necessary educational capability.

Scale of Use

Mostly using a few VM's, but with some experiments testing large clusters.