Supercomputers worldwide are crunching data with UB software

UB is making the fast-paced world of high-performance computing even faster.

That’s because supercomputing facilities worldwide are using XD Metrics on Demand (XDMoD), a software program developed at UB’s Center for Computational Research (CCR) in downtown Buffalo which improves efficiency.

As the name suggests, supercomputers are faster, more powerful and more expensive than general purpose computers. They perform specialized tasks, such as weather forecasting and particle physics, that require immense amounts of mathematical calculations.

Academics, industry and other users often must wait — sometimes weeks — to utilize these in-demand machines. To limit the backlog, the National Science Foundation (NSF) is trying to improve the efficiency of supercomputers.

One such effort is XDMoD and an open source version (Open XDMoD), which automatically identify failing or poorly performing hardware and software on supercomputers. The program has data and analytic capabilities that display historical usage trends, guide system upgrades and provide metrics to help quantify scientific impact and return on investment. Also, software developers can use XDMoD to improve code performance to maximize productivity.

Designed by CCR researchers to help ensure that NSF’s supercomputers meet the needs of the U.S. scientific community, XDMoD has developed an international following, with more than 120 supercomputer centers worldwide employing it for their operations. Users include academic and industrial supercomputing centers in Europe, Asia, Australia, the Middle East, South America and North America.

One user is the German Climate Computing Center, commonly known as DKRZ.

“We are using OpenXDMoD and its AKRR module to monitor utilization and performance of our Petaflop system called ‘Mistral.’ The AKRR is an especially useful tool for regular checks of the system on the basis of both generic benchmarks and real user applications. It allows us to verify, and moreover to easily visualize, the correct behavior of all major components in a daily fashion and at the crucial moment after a maintenance, before the system gets back to production,” said Hendryk Bockelmann, a member of the application and high performance computing department at DKRZ.

“The adoption of XDMoD by high-performance computing centers worldwide is not only an indication of the important role that supercomputers play in society, but also a reflection of the University at Buffalo’s standing as a leader within the national and international supercomputing community,” said Thomas Furlani, CCR director.

Furlani expects usage of XDMoD to grow as more organizations learn of its effectiveness. CCR also is providing support and customization services to several groups. Furlani sees this as a potential opportunity to add revenue, increase staff and boost the stature of CCR.