3 seconds sounds an awfully close to the spin up time of a disk using minimal power and AC.
How is Google’s retrieval service for non-essential data, Nearline, with its three-second retrieval latency, viable at the same cost as Amazon’s Glacier, when it uses tape with a 3-5 hour retrieval latency? Tape is cheap – but slow – and Google can't be using the stuff for Nearline. We think Nearline uses either Blu-ray …
Tuesday 28th July 2015 14:14 GMT Anonymous Coward
MAID plus analytics
Note also the very limited retrieval bandwidth. Conclusion: it's MAID with highly restricted power and I/O provisioning such that only a fraction of the disks in a given rack can be powered up at any time, and only enough bandwidth is provisioned to satisfy that fraction of the disks.
And then there's almost certainly a smart / predictive software layer that is distributing objects across the storage to maximize parallel retrieval of an object when it's requested and to minimize request contention.
Tuesday 28th July 2015 14:58 GMT Bob H
Tuesday 28th July 2015 15:04 GMT Jim O'Reilly
Google uses their scrap drives
This is my hypothesis.
Google cycles hardware on a much faster rate than most corporations. Hardware has a typical 4 year life. what do you do with the old stuff...it's starting to increase failure rates, right. Meanwhile the market for old computers is slow, so selling them isn't to good an idea.
The answer is to use older gear for bulk cold storage where it is rarely powered on. This extends drive life and reduces power drastically. And the nice thing is there's no installation costs!
The economics are compelling. There is no acquisition cost since the gear is depreciated.
The big question is what happens when they run out of old gear!
Wednesday 29th July 2015 17:04 GMT DeepStorage
Old gear won't do
While the shift the old servers to the DR site model works for SMBs that have a fixed IT staff, and therefore fixed operational costs, it doesn't work at scale. Just the extra data center space, power and cooling for servers with 1TB hard drives of 2011 vs 8TB hard drives today would make buying new kit worthwhile. Add in that those higher failure rates mean more guys in the data center doing break/fix and occasionally breaking something else and old gear isn't cost effective.
The space/power/cooling is why they turn over gear faster than average in the first place.
Wednesday 29th July 2015 04:21 GMT El Limerino
Thursday 30th July 2015 13:39 GMT SImon Hobson
Could well be using the HGST drives
Using either of the shingled drives needs filesystem support - while the Seagate drive will "work" without, it hits certain performance issues (like - stops doing anything for a while while it shuffles data) if you don't have an overlying file system that understands it.
Since Google have the brains to have their own filesystem anyway, there's just nothing at all to say they aren't using either of the shingled drives for this. Actually, for the way their storage works, the shingled drives are probably a good fit.