Hacker News new | ask | show | jobs
by LeifCarrotson 1330 days ago
> For all three, it seems their spindles, actuators, and media are starting to wear out after seven years or so of constant spinning.

Seven years is indeed a lot of revolutions if they're constantly at 5,000 RPM or more.

Are those drives really live for all that time?

I'd expect that Backblaze's business model requires lots of storage and not a lot of access/transfer - lots of write once/read never data. Their storage pods have dual 10Gb network ports, but with 60 drives in a box, each capable of 6 Gb transfer speeds, you can only full use about 4 drives at a time. Are the other 56 drives spinning at full speed all the time, or parked?

2 comments

My understanding is that repeatedly stopping and starting the drives is far harder on them then keeping them running all the time.
Generally, it is. The initial spike of electricity does the most damage. In addition, a sitting drive can eventually see the lubricants settle and stick. Once the drive is spinning, there is virtually zero friction wear. That said, a drive spun up once every several months will likely have a very long life for storage as long as the lubricants were manufactured correctly. I still have old ~500 MB drives that still work. The main advantage of having them run 24/7 is you are likely to register warnings before the drive completely dies, giving you an opportunity to decide how to deal with it. A sitting drive can simply not start up again. Plan accordingly.
Do we know any HDD that are tuned for the drives longevity? Ignoring Energy usage, speed, latency etc.
Couldn't you implement a strategy of replication and tiering to reduce wear across all drives? Some drive groups could be put to sleep and an infrequent event, like the failure of another node or disk group, would then cause these drives to spin up.

I know it’s a little different, but AWS Glacier doesn’t keep all data “hot”.

That's why I never turn my machine off: https://i.imgur.com/lHyscGS.png
I am not familiar at all with Backblaze's storage software, but I am fairly familiar with Ceph. You are not going to get 6Gbps out of a spinner. Maybe, 1.5 or 2, if it is a fairly fast drive and the data is all sequential. Plus I believe they uses SAS expanders, not multiple HBAs, so the 6Gbps bus is divided further. So they probably can saturate maybe half the drives at a time, if not more, given their bandwidth. They would probably keep them spinning, spin up is pretty power intensive, and it is physically hard on the drives.