Communications of the ACM
Why Even Google Will Embrace Cellphone Chips in the Data Center
A researcher says that assigning particular software tasks to specific types of processors would make Google's data centers more efficient, which would result in significant cost savings.
Credit: Connie Zhou/Google
University of California, San Diego professor Jason Mars spends his summers at Google's computing facilities, determining ways to make the company's data centers more efficient. His most recent research paper suggests that Google can achieve significant cost savings by using specific processors for particular software tasks.
Google’s data centers currently are designed for homogeneity, with the aim of running everything on the same hardware. However, the facilities are more diverse than intended due to machines being replaced and upgraded, and Mars and his colleagues noticed that specific applications performed better on certain processors. He says that by deliberately matching applications with chips, Google could make its entire operation up to 15-percent faster. Mars notes that even a 1-percent increase would save millions of dollars.
"It turns out that we want to embrace heterogeneity," he says. "There is a huge opportunity to build cheaper data centers that are actually higher in performance."
In line with Mars' research, companies such as Facebook are starting to support the idea that costs can be reduced with specialized chips and other low-power processors with architectures created for smartphones.
From Wired News
View Full Article
Abstracts Copyright © 2013 Information Inc., Bethesda, Maryland, USA
No entries found