Sign In

Communications of the ACM

ACM TechNews

Cutting Cost and Power Consumption For Big Data

View as: Print Mobile App Share:
chip illustration

Researchers from the Massachusetts Institute of Technology (MIT) are developing a new system that could enable flash-based servers to perform as well as RAM-based servers, but at less cost and using less energy. The researchers describe their system in "BlueDBM: An Appliance for Big Data Analytics," presented at ISCA 2015, the 42nd International Symposium on Computer Architecture.

The key to the new system is the use of field-programmable gate arrays (FPGAs), reprogrammable devices that are often used in niche applications where it is too expensive to manufacture purpose-built chips. The researchers built a prototype network of flash-based servers of interconnected FPGAs, each of which controlled two 50-GByte flash chips. The arrangement created a very fast network that enabled any server to retrieve data from any of the constituent flash drives. The researchers then tested their prototype network against a network using conventional random-access memory (RAM) and found it was just as efficient, but was less expensive and consumed less energy.

"This is not a replacement for [dynamic RAM] or anything like that, but there may be many applications that can take advantage of this new style of architecture," says MIT professor Arvind, whose group led the research. "Everybody's experimenting with different aspects of flash. We're just trying to establish another point in the design space."

From MIT News
View Full Article


Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA



No entries found