Page 7 - GCN, Oct/Nov 2016

Can the cloud rival high- performance computing?
every part of the protein interacts with all parts of the molecule.
She said Biowulf and the cloud service had similar performances, but using the cloud saved time because she didn’t have to wait for computing time on the shared Biowulf system. Instead, working with the cloud provider’s technicians, she was able to design a virtual computer struc- ture and write scripts that optimized the process and allowed her to take advantage of temporarily dedicated resources.
Using the cloud for her research meant Mills did not have to “worry about other people not having access to resources they needed,” she told the NIH Record.
“It was a virtual computer cluster that I built and I’m using, and when I’m done with it, I just shut it off,” she added. •
BY AMANDA ZIADEH
Researchers at the National Institutes of Health’s National Heart, Lung and Blood Institute wondered whether cloud computing offered any advan- tages over their high-performance computing cluster designed for biosciences.
To explore the possibilities, they did what scientists usually do — they ran an experiment.
They began by testing the use of third-party cloud providers that had been approved under the Federal Risk and Authorization Management Program.
Although the cloud was ideal for hosting the institute’s website and
cold storage, the researchers wanted to see whether the technology offered benefits for the kind of computing- intensive research they conduct.
Their test involved pitting a third- party cloud service against Biowulf, a 20,000-plus processor Linux cluster in NIH’s high-performance computing fa- cility, and they recruited Maria Mills, a postdoctoral fellow in NHLBI’s Lab- oratory of Single Molecule Biophysics, to compare the platforms. Her work involves studying how a protein inter- acts with a small molecule.
According to Mills, the lab runs simulations that are computation- ally expensive because they require quickly and repetitively calculating a massive amount of data to see how
GCN OCTOBER/NOVEMBER 2016 • GCN.COM 7