Science startup reduces compute time from 26 days to 8 hours

Print Friendly, PDF & Email

We live in troubled times, with a sharp increase in the number and intensity of terrorist attacks, acts of aggression and tensions between major international powers at an all-time high. It is no wonder that people across the globe feel increasingly uncomfortable at work, at home, and at play.

Fortunately, we also live in a time of unprecedented technological advances, and there is no shortage of people with the talent and drive to apply those technologies to improve our security and quality of life. One of the companies at the forefront of this effort is Seer Technologies, a Utah-based startup that specializes in the detection of chemical agents in the environment.

Seer’s product and the challenge of getting to market

Seer scientists spent five years developing AccuSense, a small gas chromatograph instrument that was designed to be carried around and used in the field by both first-responders and the military. A key challenge in getting the new product to market, however, was proving difficult to solve: how to separate and identify mixtures of multiple chemicals at once despite the presence of “confuser chemicals” (chemicals that are designed to hide the presence of dangerous compounds that could be used in terrorist and military attacks).

Seer had a solution based on a combination of a new chemical scrubbing process and a novel computer algorithm, but it took too long to solve. The company used MATLAB to create the algorithms that could process the chemical data and identify the various agents, even in spite of the confuser chemicals, but it took 26 days to run on the workstations they already had in their offices — far too long in an emergency situation.

Getting over the computational barrier

Seer’s problem was computational: they already had a computer solution to their problem, but they needed more computing horsepower than they had. The software package that they were using — MATLAB — natively supports the use of multiple processors at one time to speed solution, and so Seer decided to look to a high performance computing cluster to help.

The solution that they deployed was Cray’s CX1 high performance computer with 6 quad-core Intel Xeon processors (for a total of 48 cores altogether; if this is a new term for you, you might find some of the articles in HPC 101 helpful). In a form factor that is designed to fit under a desk and run off standard 120-V circuits, the CX1 can be an ideal solution for an office environment. As a startup Seer was also concerned about controlling the costs of their new deployment, and wanted something that would fit into their existing Windows-based environment without requiring new training for systems administrators. This led them to configure their cluster with Windows HPC Server 2008, a variant of the popular Windows operating systems that is designed for HPC.

From 26 days to 8 hours

Because the cluster operating system was so close to the environment in which they were already running their MATLAB computational problem, Seer didn’t have to change anything in order to take advantage of all the cores in its new cluster. The system was installed by Cray in half a day, and that first day they were running their application on the new machine in 8 hours, nearly 80 times faster than the application ran on their original desktop computer.

Take aways for your business

Seer was actually in a pretty good position to make the short leap to high performance computing. They started out with a problem that they knew was a computer problem — they were already running an application that solved the problem they needed to solve, it just didn’t run fast enough. This situation is fortunate, but does occur frequently in science-, engineering-based businesses such as consulting firms and technology startups.

If you aren’t in one of these business areas, or you don’t already have a computer-based solution for your particular business problem, finding (or developing) an application that meets your needs will be your first step. If you are in this boat, you may find What can HPC do for my business? or the articles in Next Steps a helpful resource.

Trackbacks

  1. […] See the original post:  Science startup reduces compute time from 26 days to 8 hours … […]

  2. […] Science startup reduces compute time from 26 days to 8 hours … […]

Comments

  1. Any information on why they didn’t opt for something GPU-based? Like Jacket or GPUMat ?

    I would like to know what kind of boost (if any) those offered compared to the CX1.