New Beginnings: JILA welcomes the Janus supercomputer

The Janus Supercomputer.

Image Credit
Joel Frahm and Zebula Sampedro, University of Colorado

In Roman mythology, Janus is the god of beginnings and transitions, of doors and bridges, as well as endings and time.  The aptly named Janus supercomputer at CU is bringing new opportunities in high-performance research computing to JILA. Since the fall of 2010, JILA groups directed by Andreas Becker, Mitch Begelman, Chris Greene, Ana Maria Rey, and Juri Toomre have used more than 25 million CPU hours on Janus for research in astrophysics and AMO physics.

Janus was assembled in October of 2010 at a site near 38th and Arapahoe in Boulder. Testing of the new supercomputer began in 2011 during the month of January, which was also the origin of Janus’ name. CU’s first supercomputer was specifically designed for massive parallel processing. It contains 16,415 total cores (CPUs) and can perform 814 trillion floating-point operations per second. The entire operation uses 500 kW of electricity; about 10% of that amount is used for air conditioning.

Kelvin-Helmholtz instability modeled by graduate student Greg Salvesen using the Janus supercomputer.

Image Credit
Greg Salvesen, JILA

“It’s astonishing how CU’s research computing infrastructure has gone from nearly zero two years ago to this,” said Peter Ruprecht, who recently left JILA to join the Research Computing Group. “Janus is like having 100 JILIAC clusters with a petabyte of storage along with a high-speed research network across campus that connects to these resources.”

It’s no wonder JILA’s astrophysicists and AMO theorists are united in singing Janus’ praises. During the summer of 2011, Graduate student Greg Salvesen (Begelman group) used 5 million hours on Janus to complete a fluid instability analysis for his Master’s thesis project. Using the ATHENA code, he explored Kelvin-Helmholtz (K–H) instabilities in jets emitted by black holes and other astrophysical objects. This kind of instability is the reason flags wave. It occurs when two fluids flow past each other and something “tickles” the interface, causing waves to form. Both fluids can become very turbulent. Salvesen was interested in instabilities that form at the boundary between a jet and the material around it.

“Jets are propelled incredible distances through space,” Salvesen explained. “We expect them to be subject to K-H instability, but we don’t see them break up.” For his project, he studied energy transfer due to the K–H instability and attempted to pin down the instability’s fundamental nature. The progression of instabilities between a jet (orange) and its surrounding space (blue) is shown in the figure. Salvesen’s project is the highest-resolution two-dimensional simulation ever done with ATHENA.

Salvesen puts the research that Janus made possible in perspective. “This work would have taken roughly 150 years to complete on a MacBook,” he said.

Postdoc Sean O’Neill (Begelman group) also used Janus to explore current-driven fluid instabilities. Current-driven instabilities are like K-H instabilities, but driven by flows of electrons that produce magnetic fields. They can travel at nearly the speed of light. Janus made it possible for O’Neill to study current-driven instabilities in the context of astrophysical flows, including relativistic jets and pulsar-powered systems such as the Crab Nebula. The figure  shows the evolution of a current-driven instability that can occur inside a jet traveling close to the speed of light.

Because of O’Neill’s work and Salvesen’s K–H study, the Begelman group was the single largest user of Janus during the fall of 2011. The group has already secured another 4.9 million hours on Janus for 2012.

After some initial difficulties in porting existing codes over to CU’s new supercomputer, Toomre group members are now exploiting the capabilities of Janus as well. Graduate student Kyle Augustson is running detailed simulations of fluid dynamics in the interior of F-type stars, which are a little bigger and hotter than the Sun and have an unusual near-surface shear layer. Graduate student Ben Greer is looking at helioseismic data from the Sun obtained by the Solar Dynamics Observatory to analyze real subsurface flows and compare them with simulations. Graduate student Chris Chronopoulos is investigating magneto-rotational instabilities of the Sun. And, graduate student Nick Nelson is exploring the relationship of buoyant loops in the Sun’s magnetic field to the boundary between the Sun’s convection zone and its radiative interior.

In the AMO world, theorists are using Janus to explore the quantum world. Since early 2011, for example, postdoc Yujun Wang (Greene group) has relied on Janus for research on Efimov physics.

With Janus, he and his colleagues were able to show than the Efimov effect persists with dipolar atoms in an electric field. The atoms first studied were bosons capable of occupying the same quantum state. In the Efimov effect, three such atoms can stick together in an infinite number of quantum states, even though any two cannot form a molecule.

More recently, Wang has been exploring the strange behavior of fermions, which cannot occupy the same quantum state, under the same conditions inside a Bose-Einstein condensate where Efimov trimers form. Janus is a critical part of this research, which is revealing unexpected new physics.

“Janus is so powerful,” Wang said. “Janus takes just a few hours to do the same calculations that used to take a few months at JILA.”

Postdoc Salvatore Manmana (Rey group) is reaping similar advantages from using Janus in his studies of quantum simulation, quantum magnetism, and superconductivity. He is working to predict the behavior of atoms in an experimental quantum simulation underway in the Ye and Jin groups. “Janus is helping me stay one step ahead of them,” remarks theorist Manmana.

Janus is helping graduate student Hongcheng Ni do something no one else in the Becker group has ever done: model a weakly bound helium dimer (He2) in four dimensions. Ni is studying the time it takes for information about the charge interactions between two electrons to be transferred from one to the other. So far, he has used 150,000 CPU hours on this problem.

Despite its recent contributions to JILA research, Janus isn’t competing with JILA’s cluster computers. “Janus was designed to fly through CPU-intensive parallel calculations, but it can’t handle the large-memory or disk-intensive jobs needed for such applications as quantum chemistry calculations,” Ruprecht said.

Research Topics