Backblaze, owner of 317,230 HDDs, says HDDs are “lasting longer”
- Thread starter JournalBot
- Start date
The problem with Backblaze's figures is they only have one type of workload - backup. So as much it might tell you what backup does to disks I don't think you can extrapolate this to other uses.
Upvote
38
(62
/
-24)
There's a strange oeriodicity to the graph that makes me wonder if we're seeing a methodological problem that's creating aliasing. Why would drives fail most often at "odd.5" ages, and least often at "even.5" ages? I'd expect the individual ticks to show more random noise next to their neighbors, and less of a modulated 2-year sinusoid.
Upvote
48
(50
/
-2)
Upvote
38
(43
/
-5)
Upvote
73
(76
/
-3)
Full disclosure: I’m the delivery guy for hard drives to Backblaze. I always drop the Seagate drives a couple times before taking them in.
Upvote
227
(231
/
-4)
I just need my 12TB spinners to last me about 10 years so SSDs or NVMEs can replace them eventually once pricing is reasonable (say $0.02 per GB).
Upvote
41
(44
/
-3)
Facebook has that data internally I'm sure, but they haven't published any studies that I've seen.
Upvote
32
(32
/
0)
Interesting claim. I have yet to have an SSD fail. Back when I was using HDDs, I had one fail every year or two (I typically have 3-5 drives in my main personal PC). The 7200 RPM drives seemed to fail a lot more, regardless of brand, so before making the transition to SSDs (~13 years ago?), I was sticking with those.
Obviously, my sample size is pretty small, but the high failure rate I was experiencing with HDDs (On 24/7 but definitely closer to a typical consumer workload, otherwise) was just as much reason for me to transition as performance.
Upvote
3
(19
/
-16)
Anecdotal evidence, but I've had two SSD failures total. And for those failures, I think that was a bad Sandisk lot, because I had two other drives from the same line that are still kicking since 2017 but were manufactured at a different time period. My 14-15 year old Crucial C300 is still puttering along just fine. Of course, I have some even older WD Velociraptor 300GB drives that are still working too. That said, cheaper HDDs have had a lifespan of about a decade for me and I simply wouldn't trust them to reliably backup my data past that point. In contrast, my MLC based Crucial and Samsung drives still have 95%+ health and those are all 10+ years old now.
Upvote
37
(38
/
-1)
If you look through the backblaze numbers, the failure rate for a lot of recent hard drives is more like 1-2% per year, so with modern hard drives that would be vanishingly unlikely unless you happened to buy a bunch of bad drives somehow.
This matches my own experience. We have maybe 50 hard drives in local storage at work and I think we have replaced 1 or 2 in the last 5 years. Then again we mostly just buy the ones BB says have a 1% AFR or better.
The problem with SSD isn't so much that the flash is going to fail, but rather they brick themselves due to a software bug or something on the (cheaply made) PCB fails under relentless thermal cycling.
Upvote
68
(69
/
-1)
A LOT of people used to cite Backblaze stats as a reason to not buy any kind of storage made by Seagate, and it was always annoying to me. Thankfully I don't see it as much now.
I've had 3-5 HDD's in all my computers ever (starting from the late 90's, most of which were scavanged/bought used, also I would replace my computers every year or so, so that is a lot of drives) and I think I've had ... four HDD failures, if I count the ones that worked when I took them out but a decade later was dead. Live HDD failues, as in a HDD that failed when it was in use? One.
Upvote
3
(9
/
-6)
My 320mb (megabyte) IDE IBM hard drive from 1994 still boots Win3.1 and loads Doom II just fine.
Who says older drives are unreliable?
Who says older drives are unreliable?
Upvote
68
(73
/
-5)
I've had the highest failure rates with 7200 rpm performance HDDs. Most anything 5400 rpm was slower, but cheap and reliable. I've had zero failures with upper tier or enterprise 7200 rpms drives though (e.g. RE4, IronWolf Pro, WD Gold, etc.)
Upvote
16
(17
/
-1)
One of the biggest differences between home use and datacenter use is the variations in temperature that occur at home. Drives that are in use 24/7 are going to sit in a much narrower temperature range than those that sit dormant for long periods, and then get spiked under a heavy load occasionally.
Upvote
62
(62
/
0)
Upvote
33
(33
/
0)
Fucking VERY justified. I will NEVER have another platter drive in a normal end-user computer. Do I use them in my NAS, and as external backups for my NAS? Sure, but that is a very different use case.
Upvote
37
(50
/
-13)
This matches my experience. It has been a long time since I have had a hard drive over 2 years old fail. The last one was a Seagate 3TB Barracuda, a model which was known for its excessively high failure rate forget which one), which failed at 6 years old.
I have still got a stack of 2TB and 4TB WD and Seagate drives that all function perfectly fine and have no bad sectors. They're just not really useful.
I have still got a stack of 2TB and 4TB WD and Seagate drives that all function perfectly fine and have no bad sectors. They're just not really useful.
Upvote
18
(19
/
-1)
I've had great luck with a Samsung SSD that is going on 10 years as an OS drive.
But I've steered clear of everything Sandisk since there portable SSD issues that they wouldn't acknowledge for the longest time. Then when Sandisk did acknowledge it, they claimed to fix it via firmware, but many people still reported the same failure issues.
Upvote
30
(30
/
0)
Must not have been doing this for long then. Enterprise 7200 drives were god awful about 15 years ago, while our 10k and 15k drives were at low single digit failure rates the enterprise 7200 drives were more like 10-13% with Seagate drives from HP going above 100% AFR for a few years for us.
Upvote
11
(12
/
-1)
I've still got a pair of intel X25M 160GBs striped in a machine that is booting from them, working great.
Upvote
9
(10
/
-1)
This is the kind of thing people used to say at work just before a drive let out the click of death or bled blue goop.
Upvote
31
(31
/
0)
Not sure if I'm reading it right, but it looks like you are referring to the 2013 data series that is in blue. I've skimmed the 2013 and 2021 reports that were linked to in the article.
What I've picked up is that 2013 marks the start of data collection. Looks like while some drives were older, all drives in that series were discarded (or no longer tracked) by the time they were 4.5 years old. ie by 2018 at the latest. I agree the oscillation looks odd.
Backblaze probably agree with you, as the oscillation is not there in the yellow 2021 series or the red 2025 series. I suspect the 2025 series is a look-back series, so it's collected on a different basis to the 2013 series. In theory that shouldn't make a difference, but...
PS
Some of the comments in the various Backblaze posts suggest the 2013 series shape is due to this or that infamous model of HDD that died like clockwork in year 3.5. Backblaze responded and said they tested removing these models from their 2013 data, which slightly changed the shape but not by much.
Upvote
29
(29
/
0)
When I worked at IBM, Boca Raton where the PC was made, hard disks were called, “hard files”.
This was around 2000-2001.
Upvote
-4
(7
/
-11)
I've got an original 32GB X25-E drive in use in a laptop that I ran Home Assistant on, that workload moved over to an RPI but the laptop still runs Linux. Its original life was as a log partition for a database server so it was used hard and put away wet until it got retired to my personal use =)
Upvote
7
(8
/
-1)
That, and at home PC’s are not running 24/7, so they constantly get shut down and started again.
I am always a weary when a server needs to be shut down (those with HDD at least, which becoming an endangered beast) because I’ve seen too many seemingly fine HDD not booting up.
Upvote
28
(28
/
0)
That tsunami didn't just kill a horrific number of people, it also cause reliability issues. Perhaps that is one cause to the historically higher failure rates - and now we're past that.
Upvote
17
(18
/
-1)
I've had failures with both, but since excising spinning rust to an always-on NAS that never moves and is hooked up to clean power they've lasted longer. It also helps that the NAS does more diagnostics, making upcoming failures easier to catch earlier; failure rate is still lower, but the impact of a failed drive is lower too. The SSDs, however, will just die with zero warning. It's why I run them in a RAIDZ1 configuration plus regularly back up that pool to the HDD-backed pool.
My favorite upgrade is to swap someone's 5200rpm drive for an SSD. It's like a brand-new computer. Most of those are gone now, but it was wizardry when I could tell someone to give me their laptop and by tomorrow it'll be better than new.
At a past job, long ago, I tried pushing for a project to eliminate the 1.8" HDDs with CF cards in an adapter. It would improve speed (these tablets ran full-cream WinXP) and reliability (drives died frighteningly quick). I was told that it wasn't worthwhile, because there was a maintenance contract, but I could try to speed up the tablets by slimming down the Windows XP image. Then I was told to remove the utility that used the tablet's gyroscopes to park the drive head when motion was detected because "it makes the tablets so slow".
The
Upvote
31
(32
/
-1)
True. But I do think it's a good analog for readers who are in the Data Hoarder/Home NAS market.
Upvote
25
(25
/
0)
Last edited:
Upvote
28
(28
/
0)
I know its only a sample size of like 20 for me or so, but I have yet to see an SSD that i've owned or been adjacent to fail.
How do these HDD rates compare to SSDs? I has to be higher, but does a similar dataset exist for SSDs? And yes, i am aware that you can wear out an SSD by using it enough.
How do these HDD rates compare to SSDs? I has to be higher, but does a similar dataset exist for SSDs? And yes, i am aware that you can wear out an SSD by using it enough.
Upvote
9
(9
/
0)
I had 2 SATA SSDs fail in the past year. Surprisingly, none of my 17 desktop attached drives have. And of course my back up server, which still had a 2007 dated SATA HDD that did fail, but its other 6 drives, all HDD, are performing well with various POH numbers in the 30K+ range. The one that died was over 80K.
Temps, my friend, are the real enemy. Keep them under 35C and they will reward you with long life.
Upvote
14
(14
/
0)
I bought a number of absolute bottom-of-the-barrel SSDs (DRAMless TLC SATA drives from MicroCenter's Inland brand) and those had a 10-20% failure rate over 4 years or so.
Sticking to Micron/Crucial, SanDisk/WD, SK Hynix, and Samsung, I haven't had a single failure. I definitely can't say that about hard drives. I joked about Seagate above, but I've had multiple failures from WD, too. And brands that were subsumed into those over the years.
Mine is in my PS3, where it's also still working fine, 16 years after I bought it.
Upvote
17
(17
/
0)
One unfortunately limitation is the size of the data set, the size of the drives, and the type of drives.
The last part really matters because while we had a handful of SATA SSD failures, we've not had a single SAS SSD or NVMe SSD failure. Wish I could say the same for PSUs and DIMMs, those both seem to have had some creep ups.
Upvote
5
(5
/
0)
It's old data, but I remember Techreport (RIP) doing endurance testing. https://techreport.com/review/the-ssd-endurance-experiment-theyre-all-dead/
Upvote
13
(13
/
0)
I go back to about 2015 or so with sSDs of various types and have had at least a dozen fail over that that span. IMO and only backed up by my experiences, heat and power are the enemies of SSDs. My 30+ HDDs are lasting well, and aside from a failure every year or two on them, they are mostly trouble free and the ones that failed since 2020 were all reasonably new and under warranty. Temps are the real enemy of HDDs. Keep them under 35C and all will be good. SSDs are not as easily monitored for temps and as such this can easily go undetected as a source of failure. Electrical spikes, both up and down can easily kill an SSD. Brownouts occur much more commonly than you would expect. The variance of them is usually mitigated somewhat by the utility company but they sneak in and can kill sensitive electronics. UPS are mandatory IMO.
Upvote
11
(11
/
0)
Something not mentioned in the original blog post or remarked on here is that Backblaze has improved their enclosure design over that time. They're published articles about their bespoke storage server cases and how they've evolved. That includes improved vibration isolation and temperature management. That could be a large confounding factor in them seeing decreasing failure rates in their fleet.
Upvote
72
(72
/
0)
They did some, but the last time was September of 2023. I asked about it in the comments of their 2024 year-end article, and Stephanie Doyle answered with the following:
Upvote
33
(33
/
0)
My most recent HDD failure was a Seagate drive. A Barracuda 7200.9 1TB. Died this past summer. 70K+ PoH. Some do last better than others.
Upvote
7
(7
/
0)