At the TeraGrid ’11 Conference earlier this month, John Towns delivered a keynote describing the end of the 10-year TeraGrid program and the ramp up of XSEDE, its replacement. XSEDE (EXtreme Science and Engineering Discovery Environment) will be the recipient of $121 million from the NSF over the next five years to support the new project, and Towns, who directs the Persistent Infrastructure Directorate at the NCSA and chairs the TeraGrid forum, will slip into the PI role for the new program.

XSEDE has advertised itself as less about high performance computing resources, which was the central focus of TeraGrid infrastructure, and more generally about collaborative digital environments. I spoke with Towns about what that’s going to mean for the TeraGrid user community and what they can expect under the new XSEDE regime.

The big difference with XSEDE, he said, is the shift away from TeraGrid’s technology-centric focus, which was designed to deliver primarily high-end HPC resources to researchers. With XSEDE, they’re moving to researcher-centric point of view, and in doing so, will be positioned to reach a much larger audience. At least, that’s the idea. “The intent is to create and employ an environment into which researchers can embed all of the resources they care about in order to be productive in conducting their work,” he told me.

The project will employ various distributed computing technologies, such as the XSEDE User Access Layer, to link up disparate hardware and software platforms, while providing a unified view of those resources. According to Towns, it’s quite possible most of the researchers using XSEDE will not use supercomputing resources at all. And these resources could be from outside NSF centers. So besides supers, XSEDE will encompass non-HPC systems and storage as well as database repositories, applications, and tools hosted on those machines.

How big that community could grow is still a question. The most recent assessment of the TeraGrid community pegs the number of researchers at approximately 7,000 unique users. According to Towns, that the broader availability and lower barriers to entry for this larger infrastructure could grow those numbers substantially. “I could easily imagine an order of magnitude larger number of researchers making use of it in a variety of ways,” he said.

Even though there is going to be a broader scope for XSEDE, with new capabilities and services, it doest mean they’re throwing HPC overboard. A guiding criteria for the project was to ensure a smooth transition for existing TeraGrid users, making sure there is no disruption in service. By and large, what researchers are doing from day to day should not change for them, said Towns.

For some users though, that might get a little more complicated. At the end of July, a number of TeraGrid HPC resources will no longer be available to XSEDE. These include the Big Red supercomputer at Indiana University, Queen Bee at LONI, Frost at NCAR, , Athena at NICS, NSTG at ORNL, Pople at PSC, and Ember at NCSA. Most of these are older systems, so overall not a lot of capacity will be lost, but it will reduce XSEDE’s HPC resources to just 13 high-end machines.

And even though $121 million has been allocated to the project for the next five years, it’s not clear how much NSF funding will be available to XSEDE HPC systems over that timeframe. That money is separate from the XSEDE award, and until those numbers known, it’s difficult to assess if funding will grow, shrink or stay the same, compared to TeraGrid. Even in the short term, this remains a question. The first report for XSEDE is still in the works, and until that is complete, there is no way to account for the resource providers’ contributions. Town expects it will be a little less than what they saw for TeraGrid.

The longer term prospects for HPC infrastructure under XSEDE have even more serious issues to contend with. According to Towns, the NSF has indicated they intend to stretch out the schedule for funding big HPC systems. Instead of allocating $30 million for a new supercomputer each year, as they did under the Track 2 program, these kinds of systems will only get funded every other year. That, said Towns, implies a reduction in resources in an absolute sense, which means these systems will be concentrated in just a few centers. Considering the operational lifetime of these machines is about four years, that means just two or perhaps three centers will have cutting-edge HPC available for NSF research. “There are plenty of us worried about this,” said Towns.

Despite that, he expects the XSEDE project to evolve into a much more useful platform for researchers than was TeraGrid. With the emphasis on interoperability, and a diversity of digital services and resources, rather than just big iron, Towns thinks that more types of research and a wider array of applications could be enabled. “We certainly want to see this develop as the seed for a much broader national cyberinfrastructure,” he said.