Mass spectrometers have served science well through the
years. They’re used to analyze which elements compose a sample molecule or
material. Without mass spectrometers, we wouldn’t be able to carbon-date
fossils or characterize proteins that a new drug needs to target to cure
disease. For a Palo Alto, Calif.-based scientific-instrument development
company named Varian, developing mass spectrometers has resulted in products
such as the Varian 640, which, among other uses, can determine the level of
hydrocarbon contamination in water.
The problem with spectrometers, however, is that they’re
essentially a high-maintenance load to take on if you want to build or design
one. For Varian, a spectrometer project has often required several thousand
computer hours spread over at least six weeks working on an internal pool of
processors. But, thanks to developments in cloud computing, this
labor-intensive scenario may no longer be the case.
Varian was on a tight deadline to simulate a design for a
mass spectrometer. Because of the time crunch, it turned to Cycle Computing, a
Wethersfield, Conn., company that specializes in high-performance computing and
open-source solutions in the field of cloud computing. Cycle provisioned a
fully secured cluster on the Amazon Elastic Compute Cloud, also known as Amazon
EC2.
Created by Amazon, EC2 is a Web service that provides
resizable capacity in the cloud, designed to make Web-scale computing easier
for developers. It provides complete control of computing resources and reduces
the time required to obtain and boot new server instances within minutes,
allowing users to quickly scale capacity, both up and down, as computing
requirements change. With EC2, using 100 machines for an hour costs the same as
using one machine for 100 hours—so time investment on complex computing
projects can be sharply reduced.
The bottom line: Cycle allowed Varian to reduce the time it
took to run a spectrometer simulation from several weeks to only one day’s
worth of calculation time—without any hardware purchases either. That’s because
the cluster grew to several hundred CPUs within EC2.
Varian also cut down on product development costs as only
the servers needed for the task were used. And the fully secured clusters
protected sensitive data, while the pay-per-use model removed upfront capital
costs.
Since this project, Cycle continues to increase efficiencies
for clients thanks to compute harvesting advancements. Most recently, it worked
with Purdue University on what’s called the DiaGrid project, which opened up
more than 2 million additional computer hours per month for university research
projects without any need for extra data center space, power or servers. This
increased total computing capacity on campus to more than 177 teraflops—equivalent
to the value of a $3 million supercomputer.
“The most inexpensive processors to rent in a cloud are the
ones you already own,” said Jason Stowe, CEO
of Cycle Computing. “Between harvesting un-utilized internal desktops and
servers and using cloud computing, companies can now complete a task in less
than a day, where it used to take six weeks. In industries where the ratio of
computation is still really high—like life sciences—harvested private clouds
and public clouds like EC2 will have a dramatic effect.”
