« Back

XSEDE Value Added

Report Charts XSEDE's Return on Public Investment

by Ken Chiacchia, senior science writer, Pittsburgh Supercomputing Center

The NSF-funded Extreme Science and Engineering Discovery Environment (XSEDE) delivers its essential critical services at roughly the cost of funding two independent centers and less than the cost of four independent centers, according to a paper investigating XSEDE's "value added" to NSF and the HPC community. Presented at the XSEDE15 conference in July, the report also calculated the network's value to SPs and individual researchers. XSEDE provides "one-stop access" to advanced computational resources at seven service providers (SPs), support and consulting help to tens of thousands of researchers and students.

"The return on an enterprise such as XSEDE over years or decades can be hard to track," said Craig A. Stewart of Indiana University (IU), deputy director, XSEDE Community Interaction, and the paper's lead author. "XSEDE's value added is reflected in many ways that are hard to quantify. For example, XSEDE helped researchers whose studies changed stock market regulations and made U.S. stock markets more stable. What's the value of a stock market crash prevented? A lot."

In 2013, XSEDE-powered research by Mao Ye of the University of Illinois, Urbana-Champaign, and colleagues helped convince the New York Stock Exchange and NASDAQ to change their reporting requirements to prevent unreported "odd-lot" trades from destabilizing the market.

One of the critical issues in the organization of the NSF-funded advanced cyberinfrastructure services is the service provided by XSEDE to help SPs. The investigators queried SPs on how much they would need to spend individually to replicate XSEDE's services. This survey derived a collective cost savings to the SPs of $11.7 million per year.

"We would not be able to offer our resource as a national resource without XSEDE," commented one surveyed SP representative.

"There seems to be incrementally more administrative work needed to support XSEDE than if we didn't," wrote another. "But those negatives get offset by values that aren't quantified."

Many of these non-quantified benefits are reflected in the additional user support and training that XSEDE's size enables, according to coauthor Justin Whitt of the National Institute for Computational Sciences. "We don't have, for example, a supercomputing center offering a course on MPI at the same time that another is offering a course on MPI, while nobody's offering one on OpenMP," he explains.

"We are saving money and providing these HPC services more effectively, but then there are all these other intangibles—what we call value added," says coauthor Ralph Roskies of the Pittsburgh Supercomputing Center, a co-PI in XSEDE. These include ease of transfer to systems at other SPs when a researcher discovers part-way through an allocation on one system that another is better suited for his or her project; time saved in not having to apply for separate allocations at different centers; and the value of a unified training effort in meeting the needs of both HPC newcomers and experts, as well as the increase in user productivity from having the same "look and feel" in documentation at each of the service providers.

The investigators estimated that in order to have a "return on investment" of greater than 1.0—delivering more value to researchers than XSEDE costs—the enterprise would have to provide services to users valued at $460 per year or more.

"With a standard value of roughly $100 per hour for a researcher's time, if we've saved users only five hours a year by having this unified environment we've more than returned the cost of XSEDE over multiple independent centers," Stewart says.

The study's authors were Stewart, Whitt, Roskies, Richard Knepper of IU, Richard L. Moore of the San Diego Supercomputer Center, and Timothy M. Cockerill of the Texas Advanced Computing Center. The full report can be found at http://dl.acm.org/citation.cfm?id=2792768&CFID=709509143&CFTOKEN=47819310