Comments on: Backblaze Hard Drive Stats Q2 2020

By: Mercredy Mifone

Mercredy Mifone — Tue, 10 May 2022 14:00:19 +0000

My personal experience: I have 1TB seagate since 13 years ago without difficulty though I have used it rarely. 4 WD of 1TB and 2TD was damaged but a 1TB sillicon power was good working since 6 years ago frequently. Upon my experience, I strongly reject WD because I lost so much valuable data.

By: JoeCapitalist

JoeCapitalist — Sun, 30 May 2021 19:53:31 +0000

Do you do any kind of post-mortem on the drives that fail? Drives can fail for any number of reasons (burned out motor, head crash, actuator breaks, circuit board short, etc.) it would be really interesting to get some stats on why the drives fail. Could manufacturers make their hard drives easier to repair by making it possible to just swap out the board or the motor?

By: Doug

Doug — Sat, 28 Nov 2020 19:36:54 +0000

Have you tested any WD 18TB Ultrastar HC550 drives?

By: Peter Rajdl

Peter Rajdl — Thu, 15 Oct 2020 07:02:03 +0000

In reply to John Borg. Last year, here in SA, I found the cheapest drives to get for my NAS were the Seagate 8TB external drives, and shell them. Currency fluctuations cause havoc with drive pricing in non-USA countries, especially in African countries.

By: Jeff Stone

Jeff Stone — Wed, 14 Oct 2020 22:31:59 +0000

Drive failure seems to be ruled by 2 factors: heat and vibration, in that order. I’m curious to know a few things about your drives use case, and whether you track metrics on:
-Average temperature of the drives throughout their life. Average temp, max temp, do failed drive stand out from surviving drives here?
-Average throughput load, are all drives more or less hit with the same read/write load per day? What kind of read/write hours do you see before failure? (I’d probably exclude out crib deaths)
-What about the drives position in the rack at failure? Perhaps drives are more likely to fail in the center than near the cases outer walls. Maybe the reverse is true.
-Have you done any modeling on vibration? Per drive, clusters of drives, fans, cumulative vibration caused by other surrounding pods/racks. There are videos of sysadmins yelling at a RAID array showing the read/write performance drop off a cliff. I know BackBlaze isn’t primarily concerned with throughput performance but I suspect vibrational loads can affect drive life once certain thresholds are met. Ever since I bought an Antec P183 case I’ve used PC cases that offer hard drive silicone spacers. My use case is too small but a project at your scale might be able to add vibration isolating mechanisms and derive meaningful lessons.

By: darkcg

darkcg — Thu, 08 Oct 2020 00:02:27 +0000

In reply to MrCalvin. We do not live in 1940 anymore, that's why you can't find disk bays with vibration damping features. They're simply not needed and in data centers noise is not a problem. Any modern hard drive has vibration protection technology built-in. They're built for operating at maximum vibrations a drive can produce at full load. All modern drives are also helium filled and sealed, reducing friction and hence vibrations. Not only that, hard drive firmwares monitor vibrations and adjust its operation mode on the fly. Basically, taking into account vibrations (by the way, how do you intend to quantify them, do you have a particular measuring unit?) has no meaning.

By: Kenneth_Almquist

Kenneth_Almquist — Sat, 12 Sep 2020 22:25:35 +0000

In reply to John Borg. Backblaze customers don't necessary want to wait multiple seconds (however long it takes to autoload a tape and seek to somewhere in the middle of it) for a data access, no matter how infrequently they access their data. So any use of LTO drives would require a new product offering. There are a lot of costs associated with creating a new product; it's not going to be worth it unless there is a lot of customer demand.

By: Edward Taussig

Edward Taussig — Wed, 09 Sep 2020 02:55:19 +0000

What do these stats imply for manufacturers stated MTBF ratings?
Are they still within specifications? Are any outside the ratings?

By: John Borg

John Borg — Wed, 02 Sep 2020 16:01:33 +0000

In reply to blasterman78. Would it not be trivial to RAID1 the tapes? Just copy the data onto two or more tapes hehe.