Reducing power consumption, protecting data through algorithms
Through the "unsung hero" of mathematics, Gretchen Matthews is working on ways to both make data center power usage more efficient and protect data from attacks.
Note to readers: This series of articles focuses on researchers whose work improves efficiency, addresses concerns, or offers alternative solutions to some of the pressing issues created by data centers.
The visuals of data centers can be arresting: enormous warehouses, several football fields long, sometimes backing up right against housing subdivisions.
But what’s happening inside — not just the racks and racks of graphics processing units whirring and the cooling systems working to keep the temperature in check, but the math that underpins each process — are just as important in facilitating the massive needs of the industry.
“Mathematics is really the core of so many of the technologies we use today,” said Gretchen Matthews, mathematics professor and director of the Southwest Virginia node of the Commonwealth Cyber Initiative. “It’s the unsung hero.”
Matthews has been participating in a series of data center workshops leading into the May 5 Data Center Summit event at Virginia Tech in Alexandria. Much of what Matthews and her team focus on is finding ways to protect data from degradation, loss, and malicious errors.
Back in 2024, Matthews and researcher Hiram Lopez developed a new way to store and serve data requests, an especially helpful method for places like data centers that have such enormous loads of data stored locally. This may seem like a small piece of the energy allocation puzzle, but little requests add up quickly.
“If we have to pull information from all over a center, that tends to heat things up, which leads to increased cooling requirements and energy drag,” said Matthews.
One way to avoid that is linear exact repair — what she calls a “potluck solution” and upon which she and her undergrad research team will be publishing a paper soon. Imagine if the host of a dinner party suddenly can’t cook. It’s much easier for every guest to pitch in a single dish, than have one person try to cobble the whole meal together at the last minute. Likewise, by using secure distributed computations, you can design for each server to provide a single bit instead of a full alphabet symbol and pull backup data from local servers. This provides added resilience and offers enhanced protection with built-in efficiency.
While storing data in close proximity to where it’s needed is great in terms of access, it can come with other problems. In a local model, a data center could be really reliant on a localized structure to protect information. If that data is related to critical infrastructure, losing it can be problematic, and the threats of lost data are growing each day.
“We have all this data that has been stored securely against the types of attacks that we know about now, but not for the future,” said Matthews. “I think the next thing is, what are the application-aware metrics that we can consider beyond those that we’ve adapted from other settings?”
As data centers have recently become more popular targets abroad, the need to create effective and efficient backup data storage solutions is more pressing than ever. And with news like Anthropic’s Claude Mythos model reportedly being able to find scores of widespread vulnerabilities in operating systems and web browsers, devising new ways to protect data carries as much urgency as any issue around data centers today.
“It serves as an additional call to action for post-quantum cryptography, specifically the need to accelerate migration and invest in further development of quantum-safe systems,” said Matthews.
All articles in this series
Power supply or grid stress? Both need to be solved to meet data center demand
Reducing power consumption, protecting data through algorithms
There's a power/water trade-off in data center resource allocation
Thinking small: How small language models could lessen the AI energy burden
