Backblaze HDD Report "Best" HDD
-
Backblaze have put up another Hard Drive Failure Rates blog post:
https://www.backblaze.com/blog/best-hard-drive/Discuss...
-
The data is useful, kind of, but because they don't disclose enough numbers (like how they fail over time) it isn't as useful as you might think.
-
I'm shocked by how few drives they have. I worked for a non-storage company with many, many times more drives than that!!
-
@scottalanmiller said:
The data is useful, kind of, but because they don't disclose enough numbers (like how they fail over time) it isn't as useful as you might think.
Ah, that's what was missing. I couldn't quite put my finger on it. Kind of sneaky.
-
I was quite surprised by the failure rates. I was expecting much lower rates BUT they are taking (mostly) consumer drives and beating the crap out of them in an extreme environment
-
@nadnerB said:
@scottalanmiller said:
The data is useful, kind of, but because they don't disclose enough numbers (like how they fail over time) it isn't as useful as you might think.
Ah, that's what was missing. I couldn't quite put my finger on it. Kind of sneaky.
Not so much sneaky as just "not all that useful."
-
@nadnerB said:
I was quite surprised by the failure rates. I was expecting much lower rates BUT they are taking (mostly) consumer drives and beating the crap out of them in an extreme environment
Yes, they don't have the best conditions for making drives not fail.
-
They're taking the Google approach - lots of cheap hardware and put the redundancy in the software & processes. Same thing that Tom Limoncelli advocated at Spiceworld in his talk.
Fun fact - Aaron (who used to work in support) now works for them.
-
Commodity, replaceable nodes. That's the common new approach. Nearly everyone is doing that today. It is tough for the small businesses, though, where they are lucky to have a single node. Works great when you are replacing old, legacy UNIX machines or something but when you are replacing a single, cheap node you generally can't find the budget to go to two, cheaper nodes.