Where Facebook Stores 900 Million New Photos Per Day
1sockchuck writes: Facebook faces unique storage challenges. Its users upload 900 million new images daily, most of which are only viewed for a couple of days. The social network has built specialized cold storage facilities to manage these rarely-accessed photos. Data Center Frontier goes inside this facility, providing a closer look at Facebook's newest strategy: Using thousands of Blu-Ray disks to store images, complete with a robotic retrieval system (see video demo). Others are interested as well. Sony recently acquired a Blu-Ray storage startup founded by Open Compute chairman Frank Frankovsky, which hopes to drive enterprise adoption of optical data storage.
They could just delete most of the photos after they age a bit, analyzing it with some of their AI whiz-bang software.
If anyone ever asks to see the image again, they can just show one that is "close enough" and nobody would ever know the difference.
I personally, have never posted a photo to Facebook, so I'd be OK with that.
This issue is a bit more complicated than you think.
What happens when a user wants to delete an image permanently. If it's stored on an optical disc are they going to destroy the whole disc and burn it again?
- Dan
Should have read:
You won't believe this one weird trick Facebook uses to store data!
Other than that, fascinating look at how all that data is being stored and retrieved.
I've noticed large latency for rarely used pictures in FB for over eight months now, and by large latency I mean visit the page, then come back the next day to see the next batch of > 5 year old pictures and wait another day for the final batch of ~10 years ago pictures.
After 3 months of no views, just replace them with a goatse image.
Dear God, there is more than one!?!
Left MS Windows for Linux Mint and never looked back!
Vote for Bernie in 2016!
I think he just means that optical disks don't have quite the same amount of inertia or need to do internal self-checks as HDDs, before you can actually access them. They still "spin them up", but it happens in a few ms rather than on the order of 5-15s.
Wow, they discovered HSM only 40 years after it was introduced. Amazing.
In the cloud, obvs.
systemd is Roko's Basilisk.