Flexible Chinook computer does more with less power
December 11, 2017
LJ Evans
907-474-2737
Download text and photo caption here.
A computer named Chinook has made scientific calculation at the University of Alaska Fairbanks faster and more flexible while saving more than $100,000 a year in electricity.
“Chinook is saving about $12,000 in electricity costs monthly compared to two years ago, and at the same time providing a more powerful high-performance computer system than the two it replaced,” said Gwen Bryson, manager of Research Computing Systems at the UAF Geophysical Institute.
The Penguin Computing cluster was chosen to meet UAF’s research needs because it’s big, fast and can be added to as needed — and because it eats far less energy than UAF's prior HPC systems, Bryson said.
Chinook is a community “condo” model high-performance computer cluster. It’s called that because common infrastructure elements such as racks and networking are shared and allow researchers to use their own funding to add computing segments, called nodes, to the system.
This kind of system is in some ways like owning a unit in a condominium building, sharing maintenance and associated fees with the other residents, Bryson said.
UAF scientists use Chinook to research enormous problems such as seismic wave propagation in the Earth, ozone transport in the polar regions and space plasma physics.
“By sharing common elements and pooling computing nodes, researchers gain access to a larger system than they could purchase individually and are able to focus their time on science, rather than computer administration,” Bryson said.
Modeling how earthquakes move through the ground
Carl Tape, associate professor of geophysics in the Geophysical Institute and the Department of Geosciences, conducts research on seismic tomography and seismic wave propagation with the goal of developing 3-D models of the Earth’s subsurface crust and upper mantle.
Seismic tomography is a technique for imaging the subsurface of the Earth using seismographic data about waves produced by earthquakes or explosions. These waves travel through the Earth’s layers, causing the movement of land and structures we experience when there’s an earthquake. Seismic waves can also originate from volcanic eruptions, magma movement, large landslides or large manmade explosions that give out low-frequency acoustic energy.
“The Earth is complicated,” Tape said. “Without high-performance computing, you could never convey the level of realistic details” in graphics and videos scientists use to analyze their data.
Tape said there are basically two approaches to solving huge computational problems like these. Scientists can perform their earthquake simulations at a national computing center, where there might be a couple hundred other researchers in the queue, or the computational work can be done by a local center, such as Research Computing Systems at the GI.
“The advantage to doing the computations locally is the speed of transferring massive files and the presence of experts here committed to your needs,” Tape said. “Even though I grew up in Fairbanks, I would not have applied for a job here unless this kind of high-performance computing resource was available.”
How Chinook is different
Chinook’s community condo model is a significant change in how high-performance computing resources are made available to the UA community, Bryson said.
“Previously, HPC was fully subsidized by federal agencies,” Bryson said. “Today it’s a collaboration among many UA entities, researchers and supporters.”
Researchers who contribute nodes to the system receive priority access to the cluster for a duration that aligns with the hardware warranty of the nodes they purchased. When the warranty expires, the researcher still has access to Chinook, but at a lower priority, Bryson said.
“This approach further improves the efficiency of Chinook by implementing node life-cycle management,” Bryson said. “As older nodes are retired from service, space is created in the computer racks for new nodes to be brought online, but only if needed and purchased.”
Chinook is significantly more capable than previous, legacy HPC systems at UAF, Bryson said.
“A job executed on Chinook runs in less than half the time as the same job executed on the university’s legacy HPC systems,” Bryson said.
Chinook came online in January 2016. Funding for its original installation and a recent expansion came from the M.J. Murdock Charitable Trust in a partnership with the Geophysical Institute, UAF vice chancellor for research, UAF International Arctic Research Center and UAF IDeA Network of Biomedical Research Excellence.
More than 150 researchers and students use Chinook, consistently employing over 80 percent of the system, Bryson said.
“In HPC terms, that indicates a very busy and well-utilized system,” she said.
The new system seems to be meeting researchers’ needs.
“I’ve been very happy so far,” Tape said, an attitude that Bryson said is generally echoed by the other scientists using Chinook.
“We can expand Chinook as the research community dictates and university funding or agency can support,” Bryson said. “This flexibility, along with the significant savings in the costs to power the system, is the big plus.”