Advanced Computing in the Age of AI | Friday, March 29, 2024

Is a Zero-Watt Solution Worth the Wait? 

Today's businesses move at a rapid pace. Datacenters need to keep up with that pace. However, a datacenter will use up energy to instantaneously access a file. SGI proposes that waiting a few seconds can save thousands of watts of power. 

Today's businesses move at a rapid pace. Datacenters need to keep up with that pace. However, a datacenter will use up energy to instantaneously access a file. SGI proposes that waiting a few seconds can save thousands of watts of power.

One of the biggest issues pertaining to datacenter power consumption is the actual disk drives a center houses. Five watts of power is used each time a drive is accessed. If there are 100,000, or even a million of these in a datacenter, that's between half a megawatt to five megawatts of power being used if all of the drives are running at the same time.

"One question SGI's customers have is how many of these drives are actually being used at any given time?" said Dr. Eng Lim Goh, SGI CTO.

If the answer is a low percentage of drives are spinning, should the majority of the drives not in use be powered down?

SGI has a zero-watt disk solution that manages the output of drives. It can power down any drive that currently is not in use, reducing costs and power consumption in a big data storage center. Booting up a sleeping drive can take time that some may not have the patience for.

In a demonstration, Lim shows the difference between accessing a file from both an active driver and a powered-down one that uses zero watts of energy. The active driver's file came up immediately, whereas the other file took 14.6 seconds to access.

"The question is whether this delay is tolerable to the user," Goh said. "If you can wait 15-20 seconds, you can keep your files on a zero-watt disk."

Related Articles:

Green Revolution Powered by (Mineral) Oil

Earth Day Special: Making Datacenters Greener

Earth Day Special: 10 Reasons Why You'll be Buried Alive

EnterpriseAI