MIT’s New Server Network Design Ditches RAM To Cut Big Data Costs

Researchers at MIT have built a server network that ditches RAM in favour of flash storage, yet extracting equivalent performance as more traditional servers for several Big Data applications.

Computers like to store data they’re working on in RAM where the CPU can retrieve data tens of thousands of times faster than a computer’s hard drive. However, RAM is very expensive.

The MIT researchers have been able to get flash drives pre-process some of the data instead of dumping all the work onto the CPUs and while there’s still a speed gap, it’s negligible.

Flash memory is 10X cheaper than RAM and 10X more power efficient and the researchers found that if traditional servers pull data from the disk even 5% of the time, their performance falls to a level comparable with flash.

Still, the new style of computing architecture isn’t a replacement for DRAM that found in devices, so don’t expect your smartphone or PC to go RAM-less anytime soon.

MIT’s technique however can help with database-heavy tasks such as ranking web pages which does translate to faster cloud services at lower costs and massive energy savings.

Leave a Reply

Sign Up for NextBigWhat Newsletter

Daily.