The amount of information organizations must process at the edge has exploded. This is especially true for federal agencies and the military, which generate enormous quantities of data from mobile devices and sensors in equipment, buildings, ships, aircraft, and more.
Finding effective ways to manage, use, and protect that data is challenging. But there’s an effective and cost-efficient solution. The combination of neuromorphic processing and self-searching computational storage can enable organizations to quickly process vast troves of edge data.
The Edge Data Dilemma
Edge data can provide insights that enables more effective core missions. Trouble is, the compute and network infrastructure needed to handle that data hasn’t kept up. Organizations lack the compute power to process the data at the edge, and they lack the network bandwidth to transmit the data to a centralized location where they have processing power.
Traditional computing technology takes up too much space and generates too much heat to be useful at the edge. Traditional network technology can’t move extremely large quantities of data over long distances at useful speeds. As one example, the average U.S. Navy ship produces petabytes of data from crew, operational systems, weapons systems, and communications. For many use cases, that data can’t be processed till the ship has docked.
The Cyber Advantage
Agencies must not only find effective ways to manage data, they also need to protect their assets from cyber threats. Today, cybersecurity teams must sift through enormous quantities of data when responding to cyberattacks. To uncover anomalies and home in on root causes, they need to search large datasets from access logs and security information and event management (SIEM) systems. They also need to complete that task in as near real time as possible to prevent a mission-disrupting cyber breach. But to date, they’ve lacked an effective compute and storage solution to achieve that goal at the edge.
A new report from Cyberedge found that 68% of government agencies faced a cyberattack in 2021, underlying the need for agencies to find innovative solutions for data protection in case of an attack. Active response capacity can be critical when responding to a cyber incident, substantially reducing cyber risk and protecting the mission by quickly finding the data and alerting analysts in real time.
The Power of Neuromorphic Processing
It would be helpful if computers functioned more like the human brain. A human can look at a field of thousands of yellow flowers and instantly spot the single red flower. A computer needs to process each flower individually until it can find the anomaly.
That’s because the brain has been fine-tuned over eons of evolution to perform specific tasks very well. And it does so while consuming remarkably little energy.
But what if, like the brain, a computer could perform a specific task very quickly while requiring very little power? That’s the promise of a neuromorphic processor – essentially, a computer modeled after systems in the brain.
Here’s how neuromorphic processing can transform cyber risk at the edge. Start with a neuromorphic processing unit (NPU) built on a high-end field-programmable gate array (FPGA) integrated circuit customized to accelerate key workloads. Add a few dozen terabytes of local SSD storage. The result is an NPU-based, self-searching storage appliance that can perform extremely fast searches of very large datasets – at the edge and at very low power.
Just how quickly can NPU technology search a large dataset? Combine multiple NPU appliances in a rack, and you can search 1 PB of data in about 12 minutes. To achieve that result with traditional technology, you’d need 62 server racks – and a very large budget. In testing, the NPU appliance rack requires 84% lower CapEx, 99% lower OpEx, and 99% less power.
Imagine the advantage of searching a petabyte of data in minutes when responding to a situation like the Sunburst hack. Affecting at least 200 organizations–including government departments such as Defense, Homeland Security, Treasury, Commerce, and Justice–the Sunburst hack began around March 2020 but wasn’t discovered until December 2020. Agencies had to search at least nine months of data to determine where breaches occurred, current breach activity, and which systems, networks, and data were affected.
Neuromorphic processing and self-searching storage can slash incident response times in situations like this. That can save costs, accelerate incident resolutions, and reduce cyber risk.
Making the Use Case for NPU Appliances
The NPU search technology was developed in collaboration with Sandia National Laboratories, an R&D lab of the Department of Energy. Today Sandia is actively using multiple NPU systems for cyber defense and other use cases.
One compelling aspect of am NPU appliance is that it can help organizations comply with President Biden’s May 2021 Executive Order on Improving the Nation’s Cybersecurity. In response to the order, the Office of Management and Budget issued a directive requiring agencies to retain 12 months of active data storage and 18 months of cold data storage. For many agencies, that presents a serious budgetary challenge. Am NPU appliance can make such data retention cost-effective.
What’s more, deployment of NPU appliance storage requires no changes to an organization’s current IT infrastructure or cyber defenses. The appliance simply sits alongside existing hardware and cybersecurity solutions. Searching of large datasets occurs at the edge. Any small quantities of relevant data identified can quickly and easily be transmitted for centralized analysis.
There are other potential use cases for an NPU appliance. For instance, one Fortune 50 company used the technology for data labeling to train a machine learning algorithm. The organization reduced the time required from one month to 22 minutes. In the meantime, for federal agencies and the military, neuromorphic processing and self-searching storage is an achievable, cost-effective solution for protecting sensitive data and slashing cyber risk at the edge.
About the author: David Follett is the founder and CEO of Lewis Rhodes Labs. David is a senior technology executive with 30 years of experience in semiconductors, optics, computer architecture and neuroscience. He started his career at Bell Labs Murray Hill and was the founder and CEO of GigaNet, a networking start-up that invented virtualized interfaces, ultimately evolving into Infiniband.
Neuron Research Opens Doors for Neuromorphic Computing
Bridging the Gaps in Edge Computing