Machine learning can make use of sensor data, and distributed cooling can boost efficiency, say researchers at CIRC / McMaster University

The Computing Infrastructure Research Center (CIRC) at McMaster University, Ontario is carrying out research to improve data centers, according to two of the center’s leading research engineers.

Ghada Badawy and Souvik Pal discussed how machine learning can make use of sensor data, and described a distributed cooling system that places cooling units at rack level, in an interview at the DCD>Canada 4.0 in Toronto in December 2017.

”We’re building wireless sensor nodes to monitor temperature pressure vibration and sound in a data center,” said Badawy. “We’re not just collecting the information and sending alarms - we’ve taken it to the next step where we have built some machine learning algorithms that will predict what’s going to happen next.”

While many data centers are heavily instrumented, some ninety-nine percent of data isn’t used right now, she said. Machine learning could sift through more of that data.

Meanwhile, Pal wants to get cooling to where it is most needed: “We’re working on a whole new architecture which can deliver cooling at the rack level. One example is a cooling unit that looks like a 2U server, that you can mount in any 2U space.” It can be retrofitted in older data centers to eliminate hot spots.”

The CIRC may be the only academic research institute entirely devoted to the innovations in data centers which support the cloud, said the pair. It has access to the university’s engineering staff and resources, and partners with industry.