Skip to main content

Nimble and Thrifty Data Centers

Data centers will 'see' into the future thanks to research led by computer science professor Tajana Simunic Rosing
Data centers will "see" into the future thanks to research led by computer science professor Tajana Simunic Rosing

Data centers are big and bulky on the outside, but computer scientists at the Jacobs School are making them agile and flexible on the inside.

Tomorrow's data centers will be able to "see" into the future thanks to algorithms that predict future conditions based on real-time data collected by embedded sensors. With these predictions, data centers can proactively direct incoming jobs from one server to another, turn fans on and off, switch power management strategies, and increase cooling - all in real time.

"The big picture is that we want to find a better way to manage data center energy costs," said Tajana Simunic Rosing, the computer science professor leading the projects, which are tied to UCSD's Center for Networked Systems.

Temperature prediction algorithms, for example, reduced costly, performance-busting hot spots on servers by as much as 80 percent. The team's workload prediction algorithms guide power management strategies for servers that can provide energy savings of up to 70 percent.

Rosing and her graduate students are now working to integrate the temperature prediction and workload prediction systems, and to scale up their implementations from server racks to full-scale data centers.

This summer, Rosing was tapped to lead the Large-Scale Systems branch of a $3 million multi-university initiative. This project is focused on a forward-looking computing landscape in which applications run in distributed form on platforms that mesh data centers with broad classes of mobile devices that are surrounded by large swarms of sensors. This Multi-Scale Systems Center (MuSyC) is funded by the Semiconductor Research Corporation's Focus Center Research Program.

Print Article