AWS Public Sector Blog

Needle in a haystack: How the Pacific Northwest National Laboratory leverages the cloud to power its national security image similarity solution to better serve its customers

Tasking someone to find a singular image in a sea of millions of captured images would be like telling them to find a needle in a haystack. Yet, this is one project for which Ralph Perko is developing a solution. Perko is a lead senior software engineer at the Pacific Northwest National Laboratory (PNNL) in the state of Washington. In his role, he is responsible for taking the lab’s research and creating applied, scalable solutions to support PNNL’s customers’ missions.

Why would you buy hardware?

As one of the United States Department of Energy’s national laboratories, PNNL is cloud agnostic. Over the years, Perko has been an advocate for PNNL to adopt the cloud in order to advance its research efforts and better serve the lab’s customers. He recalls about 5 years ago, he wrote a white paper about how a particular PNNL program could go about adopting the cloud. “Things came to a head in 2015, when we had new deployments we needed to get out, but we weren’t able to meet our sponsors’ needs internally.” Perko explains that the lab’s inability to deliver the solutions could be attributed to inadequate internal systems: “Internal IT infrastructure at PNNL wasn’t built to support big data as it exists today. Internally, we were running many servers. But, we were beginning to hit critical bottlenecks in our networking capacity and ability to get in the hardware we needed.” While the existing IT infrastructure worked well for some projects, with changing customer demands, PNNL had to look for a solution that delivered more agility and flexibility in order to meet their sponsors’ needs.

Faced with this roadblock, Perko and team made a decision that would change the way the lab not only conducted research, but also how they could ultimately collaborate with and deliver solutions to customers. “We made a decision to go to the cloud,” Perko says. The choice to migrate workloads to the cloud proved to be an invaluable decision according to Perko, “It’s been critical for us to migrate, and many of our sponsors are also moving to the cloud because of the Cloud First government mandate. In today’s world, why would you buy hardware?”

Perko no longer has to reference his white paper written nearly 5 years ago. Now, he says, “We do a lot of our research and development on the cloud, and solutions are deployed in our customers’ cloud environments. It makes it easier all around from a policy perspective.” Moving quickly to research, build, and deploy are some of the main benefits of the cloud, according to Perko. He says benefits range from: “Time to market. Costs reduction for services and ownership. The ripple effect of how much effort and how much of people’s time has to go into hardware maintenance… the cloud has changed all of this.”

Finding a solution in the cloud

With the cloud, Perko’s goal to build an image similarity solution became a manageable task. “Machine learning, deep learning, our ability to spin up various resources for this project on AWS, all of this helps decrease our time to market, and it helps us, as software engineers, to engage at an early point,” he says.

In order to build the solution, Perko and team have relied on tools such as AWS Lambda. He notes, “There’s a proliferation of imagery across domains in research and in national security. One of the solutions we came up with was image feature similarity. We extract features from images, and we index them, and then we are able to search on them. We use various AWS services to achieve this, like Lambda. Lambda far outperformed what we could get out of an Amazon Elastic Compute Cloud (Amazon EC2) instance, in terms of the time it takes to extract the features of these images.” On his experience with Lambda, Perko states, “we love working with serverless, and we use it whenever we can.”

When thinking of a real world example of the image similarity solution, Perko cites disaster recovery. “Imagery can be used to help identify areas in need of assistance, for example identifying the location of a flooded area. Being able to provide a platform where they can search for a particular image feature is powerful,” says Perko.

PNNL was able to provide a solution that made the seemingly impossible, possible. Now, with the power of the cloud, finding the proverbial needle in the haystack isn’t such a daunting task.