Slower Disks will save the world (or $$) by Peter Judge
Open Compute emerged as a way to have open discussion about making data centers more efficient. It may have been partly in response to criticism from Greenpeace over Facebook’s use of coal-fired electricity, and it certainly contrasts with Facebook’s normal attitude, which sees sharing as something for its customers, not itself.
Be al that as it may, Open Compute set up a Facebook group (obviously) and a blog, where members can access the information that Facebook shared - special server builds that reduce materials and energy use - and suggest their own ideas.
In that forum, Facebook engineer Eran Tal has made a suggestion that has caught many people’s imagination. A lot of data is “ciold” - rarely accessed, but not yet ripe for archiving and forgetting about. While cold servers reduce their power use, hard drives holding cold data generally carry on spinning at full speed, because turning the hard drive off would lead to unacceptably high read times, on the rare occasions when users need that “cold” data.
Tal’s idea is that, instead of turning those drives off, they could be run at a slower speed. Read times would go up slightly, but only on comparatively rare occasions, and the overall power use of the storage pool could be reduced substantially.
Saving 3-5W of power per hard disk could add up to many kW or even 1MW in large data centers.
The drawback is that hard drives aren’t currently made that way. Some of them have a slow-speed standby mode, but they don’t read and write at that speed.
The beauty of a forum like Open Compute is that ideas like this get scrutinised and kicked about by people who understand the possibilities. A Western Digital employee called Brandon Smith has showed up, suggesting better choices for the low-speed operation, and the whole idea has become more like a practical suggestion, than just a bit of blue sky thinking.
How likely is it to become reality? That’s too early to say. But if a few people at Facebook and elsewhere - with big HDD purchase budgets - take this up as a campaign, drive manufacturers could find its worth their while to implement it.
And if nothing happens immediately, we will all know about it. The idea is published, and the rest of us can follow up with Tal, or with others, or even pick it up and run with it ourselves.
This is exactly the kind of result which Open Compute was created for.