Hard Drive Stats for Q1 2017 | Hacker News Reader

Hard Drive Stats for Q1 2017

backblaze.com

62 pointsingve4 comments

Something that got my attention is that they compare Annualized Failure Rate as-is, i.e. "whichever is higher". From statistics viewpoint, that sounds not quite right, because that number may happen just as a random fluke.

Instead, every statistician would try to disprove the null-hypothesis with a predefined p-value. P-value is not a holy grail, but it's better than just comparing means.

I can also attest that the Seagate STX000DM000 is not a good drive. Out of 4 ST2000DM000 drives 1 had to be RMA'd and another is showing potential signs of failure (checksum errors in ZFS). Granted it's a small sample size but I recognized that model number immediately since I have been dealing with them a lot recently.

Personally I've found that the 40-45$ refurbished 2TB enterprised hitachi drives on Amazon are more reliable. Since I'm running in a raidz3 config I'm not terribly afraid of any single disk failure

If any Backblaze employees are reading this, can you give any insight into >8TB drives and when you expect to start deploying those? Have you tested them at all for longevity, or are they still too expensive?

Brian from Backblaze here.

We are deploying a few 10 TByte drives within a couple of weeks, so you should see some stats on them the very next quarterly drive stats (3 months from now).

We start by deploying one full "pod" (60 drives) plus an entire "tome" (20 drives spread across 20 separate machines in our datacenter which makes up a "vault" all working together) to sanity check this is a good idea. So there will be 80 drives minimum plus any spares we swap in.

If it all checks out (like they don't fail catastrophically) then they get entered into the price spreadsheet. In other words, each month when seeing which drives we should purchase, we take into account the larger density means we have to rent less data center space but the drives cost more, plus take into account the best known failure rates and what the warranty is (some drives are not in warranty so we will pay cash for the replacements), and then get quotes from our drive vendors and an answer pops out as to the least expensive drive (over a 5 year lifetime of ownership).

> are they still too expensive?

SO FAR the answer has been "too expensive" but it is worth pre-qualifying the drives as "not terrible in our environment" in case that changes. The moment we decide they have a reasonable chance of being the most cost effective we might need to buy several million dollars worth at once. And that can happen suddenly for reasons totally mysterious to us. It might be the manufacturers have problems producing enough of them, and then suddenly get the manufacturing stabilized and can produce enough. Or maybe they are just gouging the early adopters? Either way, there are SUDDEN price fluctuations we don't understand.

Reply on news.ycombinator.com