DCSL at NCAR

Distributed Climate Simulation Laboratory

Transporting gigabytes of data and performing distributed visualization between supercomputing centers.

"DCSL may be a prototype of what the user community will be doing routinely by the end of the decade."
--Bill Buzbee, NCAR Scientific Computing Division director

Beyond the Internet

NCAR and The Distributed Climate Simulation Laboratory (DCSL) is a collaborative program of the National Center for Atmospheric Research (NCAR), the Pittsburgh Supercomputing Center (PSC), and the San Diego Supercomputer Center (SDSC). The purpose of the DCSL is to evaluate the degree to which the National Science Foundations very-high-speed Backbone Network Service (vBNS) can support distributed supercomputer applications and the transport of large datasets.

The vBNS, which replaces the old NSFNET backbone of the Internet, interconnects NCAR, PSC, SDSC, the Cornell Theory Center, and the National Center for Supercomputing Applications. It can transfer information at the rate of 155 megabits per second, compared to 45 mbps for the old backbone. Using the vBNS, the DCSL focuses on two projects: (1) distributing the NCAR Climate System Model between NCAR and SDSC, and (2) transporting and visualizing datasets from the Grand Challenge Turbulence Project between NCAR and PSC. Specific program objectives are:

To assess vBNS stability and performance for the exchange of data between distributed model components
To assess vBNS performance in transporting large datasets
To develop and demonstrate high-performance distributed visualization
To develop software for the management of distributed jobs

Coupling a climate system model over a high-speed network

The Distributed Climate System Model (DCSM) project is a joint effort between NCAR and SDSC. The project is assessing the capabilities of the vBNS for supporting data communications between model components of a coupled climate system simulation that are executing in parallel on high-speed computers at NCAR and SDSC. Data interchange between model components is in the form of two-dimensional fluxes. These fluxes are exchanged over the vBNS using Parallel Virtual Machine (PVM) message passing, optimized for the high-speed network. The simulation is instrumented using HERMES, a data-capture and visualization package. The fluxes and host performance metrics are accessible in graphical form, which allows researchers to monitor model performance and host performance in real time.

A Grand and Turbulent Effort

NCAR and PSC are exploring the potential of the vBNS to transport and visualize extremely large datasets (25 gigabytes-1 terabyte) from the U.S. High-Performance Computing and Communications (HPCC) Grand Challenge Turbulence Project. Project researchers are studying highly turbulent flows, typically constrained by rotation and stratification, that play crucial roles in determining the large-scale dynamics within oceans, atmospheres and stars.

Because turbulence research produces huge volumes of output, visualization analysis is fundamental: scientists must be able to see the formation of coherent structures in the three-dimensional fluid flows they are studying. To visualize turbulence datasets in a wide-area environment has traditionally been difficult because of the sheer quantity of information, low bandwidth between centers and universities, and a lack of infrastructure supporting such activities. The DCSL provides substantial computational, networking, storage, and visualization support to address these problems.

Return to Exhibit Home Page