Feature - Massively parallel: NERSC, OSG and HPC

Franklin, named for scientist Benjamin Franklin, is the third and largest parallel resource that NERSC has connected to the Open Science Grid. Image courtesy of wikipedia

The Open Science Grid has long recognized that there is strength in numbers. This can mean large numbers of computing elements, a growing number and diversity of users, or the increasingly wealth and diversity of available resources.

A new facet in the OSG crown is a 20,000-core Cray XT4 parallel computer at the National Energy Research Scientific Computing Center. Called Franklin, it joined the list of OSG resources in 2007.

NERSC was the first OSG member to bring large-scale parallelism to the OSG and is the largest OSG resource, with approximately 25,000 cores in service.

NERSC is also the most diverse OSG resource provider, offering five different architectures. Such capabilities expand the scope and impact of OSG-enabled science.

Grids for data, jobs and effective science

While the definition of grid can sometimes be vague, the NERSC facility takes a concrete view that grid services are about managing data, managing jobs and improving the effectiveness of computational science. Such needs are perennial and cover high energy physics to climate, and QCD to astrophysics.

In 2007, climate simulations dominated OSG allocation of parallel computing time at NERSC; in 2008 parallel chemistry codes—like the one used for this simulation of the double photoionization of a hydrogen molecule—are seen as a growth area. Image courtesy of courtesy of Wim Vanroose

Grids also become part of the lifecycle of HPC application software, which often begins with singleton runs of recently developed codes, and then matures to a point where the bottleneck to progress is managing the large volume of runs and data. It is here that grid services and the OSG software stack have a lot to offer.

Grid services also provide more advanced mechanisms for initiating jobs based on web services rather than on hand-crafted batch scripts, helping to make scientific application software easier to use, and thus more useful.

Grid and web services are an excellent conduit for growing scientific collaborations. As more projects produce more high quality data sets, such services are increasingly required to further scientific insight.

Challenges past and present

The integration of NERSC into OSG has presented some interesting challenges, including software changes and tests, and some reconsideration of the means by which the usage policy and security requirements are enforced at NERSC. NERSC and LBNL staff are actively finding a middle ground between highly usable and well regulated deployment of HPC resources to grid users.

- D. Skinner, W. Kramer, J. Hules and D. Olson, NERSC