New Faces profiles rising researchers and new leaders who will drive scientific computing in the days ahead.

CONTACT

Hank Childs

Laurence Berkeley National Laboratory

This email address is being protected from spambots. You need JavaScript enabled to view it.

Personal web page

RELATED LINKS

LBNL Visualization Group
 
VisIt Project website

But he and fellow visualization researchers fear looming logjams, in both input/output and data movement, as supercomputing advances in about five years from the current petascale level of a million billion floating point operations a second (flops) to the exaflops range.

"If somebody hands you a petabyte or, in the future, an exabyte, how do you load that much data from disk, apply an algorithm to it and produce a result?" Childs asks. "The second challenge is complexity. You only have about a million pixels on the screen – not many more than in your eye – so you have to do a million-to-one reduction of which data points make it onto the screen."

Such pare-downs raise questions about the remaining data's integrity, Childs says. "And even if this data integrity issue is managed, we may still produce images so complex that they overwhelm what our human visual processing system can understand. So we're not doing anything meaningful."

Exascale computing will pose even more special challenges than that, although operators would seemingly be in a better-than-ever driver's seat. "One bottleneck will have to do with data movement between (processor) nodes and even from primary memory to cache to registers. And it will be a real problem if we have to store it or read it back. All that data movement will take too much time. Secondly, it will take power – the central issue at the exascale."

Childs cites Argonne National Laboratory's Mira, an IBM Blue Gene/Q system that uses about 4 megawatts of electricity to calculate at a little more than 10 petaflops. That's much more efficient than many earlier computers but still energy-hungry when taken to exascale. Most experts believe 20 megawatts is the limit for computer power consumption, Childs says. So an exascale version of Mira would be 100 times faster but could fuel that speed with only five times more power.

"In every part of the simulation process, people will be asking 'How can we do this in a power-efficient way?' And visualization and analysis is no exception. We also have to rethink how we will do things. For us, it really comes down to minimizing data movement, because that costs power."

Breaking with tradition

In visualization, that will mean abandoning the traditional, time-consuming practice of regularly storing simulation snapshots – at full resolution – on disk, then later reading them back on separate resources to apply visualization algorithms and create images that help interpret data.

Though he's only starting his five-year research program, Childs expects his "Data Exploration at the Exascale" Early Career project will focus on creating techniques that avoid regularly saving the full simulation for visualization and analysis. Running at exascale will make his task more complicated: Like the simulation itself, data processing will have to be executed in a billion-way concurrent environment while minimizing information flow to trim power costs.

To do that, visualization and analysis researchers have seized on in situ processing: running visualization and analysis routines concurrent with the simulation and operating directly on the data. In situ visualization eliminates the need to record simulation data but works best when application scientists know at the start what they want to study. Without that, their only choices are to pause the simulation and explore the data or accept that the data will be lost forever.

Childs' approach is to reduce the data using in situ techniques and then record it. "Of course, how to reduce the data while keeping the portions scientists want to study and also maintaining the data's integrity is a significant challenge."

Childs' project summary uses a culinary analogy:"When a cook wants to produce a specific desserthe consults his cookbook and learns what steps andingredients are necessary and how long it takes to bake...I'm hoping to create a cookbook for application scientiststo say 'I'm interested in doing this sort of analysisand have these data, and I know this will be the righttechnique to meet my power, data and integritycosts.'"

Visionary path

Childs, who turned 36 this summer, first wanted to be a high school math teacher. Early in his undergraduate years at the University of California, Davis, he bowed to a request from his father, a Ph.D. in electrical engineering, to "take one programming course," Childs recalls. "My dad was right. I found I was really good at it. So instead I became a computer scientist."

After graduating in 1999, the super-organized Childs went shopping for a job and a Ph.D. program and found both in visualization, choosing to study at Davis and considering three positions at Livermore. He took the one in visualization because "I just thought I would work well with the guy who would be my boss."

At Livermore, where Childs remained until moving to Berkeley Lab in 2009, he worked on designing visualization software that could work directly on large dataset computers. He became the main architect of what he and his colleagues named VisIt. Childs' Ph.D. thesis was on a related subject: "An Analysis Framework Addressing the Scale and Legibility of Large Scientific Data Sets." Fortuitously, both coincided with the dawn of exploding interest in high-performance computing.

"How amazing for me," he says. "I went and got a job just because I wanted to get a Ph.D., and I stumbled into an amazing situation at the forefront of HPC at the intersection of HPC and vis."

Sidebar: VisIt: More than pretty pictures

In 2000, Lawrence Livermore National Laboratory researchers set out run a simulation and allow visualization and analysis, all on the same computer. No small feat.

"The simulations were producing data sets so large that they had to be run on massive parallel-processing supercomputers," recalls Hank Childs, then a visualization researcher at the lab. Meanwhile, "the visualization and analysis work that followed was done on smaller, serially computing desktop machines," but the data were exceeding the desktops' capacity.

With support from the Department of Energy's Advanced Simulation and Computing program, he and colleagues from Livermore and other institutions created VisIt. With its open-source code, VisIt eliminates the need to move data from one machine to others. Its client-server design still allows portability to the desktop when needed – for instance, when experts want to enhance visualizations by employing graphics cards like those in video games.

VisIt's name is somewhat misleading because "it's really for a lot more than visualization," says Childs, the software's chief architect. "It makes a lot of pretty pictures, but some people also use visualization to debug what's going on in their software." VisIt enlists several techniques that let users see what is in their datasets.

In 2006, DOE's Visualization and Analytics Center for Enabling Technologies made VisIt the center of its strategy to deliver a production-capable visualization and analysis tool. Since then, universities and other national laboratories have taken up its continuing development and funding.

VisIt received a 2005 R&D 100 award from R&D Magazine. And in 2009, it was the first non-simulation code to be certified for scalability – the capacity to sustain performance levels as work demands and processor numbers increase – under an ASCR program for code improvement and scaling.

VisIt has been downloaded hundreds of thousands of times, Childs says – a separate measure of success.