SSD reliability is only slightly better than HDD, Backblaze says

[Alfonso Maruccia]

In context: Backblaze is a California-based company dealing with cloud storage and data backup services. Every year, the organization provides some interesting reliability data about the large fleet of storage units employed in its five data centers around the world.

For the first time, Backblaze's latest report on storage drive reliability is focusing on Solid State Drives (SSD) rather than HDD units alone. The company started using SSDs in the fourth quarter of 2018, employing the NAND Flash-based units as boot drives rather than data-storing drives. Backblaze uses consumer-grade drives, providing Annualized Failure Rate (AFR) information about 13 different models from five different manufacturers.

The 2022 Drive States review is based on data recorded from 2,906 SSD boot units, Backblaze states, and it is essentially confirming what the company was saying in its 2022 mid-year report. SSDs are more reliable than HDDs, Backblaze says, as they show a lower AFR rate (0.98%) compared to HDDs (1.64%).

The fact that the difference in reliability level isn't exactly staggering (0.66% AFR) is rather surprising, however, as SSDs are essentially just moving electrons through memory chips while hard drives have to deal with a complex (and failure-prone) mechanism employing spinning platters and extremely sensitive read/write magnetic heads.

The reasons behind failing drives aren't known, as only an SSD manufacturer would have the equipment needed to make a reliable diagnose. For 2022, Backblaze says that seven of the 13 drive models had no failure at all. Six of those seven models had a limited number of "drive days" (less than 10,000), the company concedes, meaning that there is not enough data to make a reliable projection about their failure rates.

An interesting tidbit about Backblaze's report is that the company hasn't used a single SSD unit made by Samsung, which is a major player in the SSD consumer market. One possible explanation is that Samsung drives aren't cheap, and Backblaze is essentially using the cheapest drives they can buy in bulk quantities.

Aside from the small difference in reliability between HDDs and SSDs, another (unpleasant) surprise coming from the latest Backblaze report is how useless SMART reporting can be when it comes to solid-state disks.

The Self-Monitoring, Analysis, and Reporting Technology (SMART) attributes have been historically used to report and analyze HDD's general health, but SSD manufacturers are using those stats inconsistently and in a non-standardized way. This means that you can't make a valid comparison between different drive models based on SMART stats alone.

Permalink to story.

https://www.techspot.com/news/97909-ssd-reliability-only-slightly-better-than-hdd-backblaze.html

 

[ScottSoapbox]

60% of the failures is still significantly less if not the 1% most of us would have probably guessed.

It is worth noting the boot drives are used continuously where the data drives only when a backup is occurring.

 

[KofeViR]

I have a 3TB WD Green running on year 12 for 24/7 usage, with headparking disabled tho. My experience with WD is great in general.

 

[yRaz]

60% of the failures is still significantly less if not the 1% most of us would have probably guessed.

It is worth noting the boot drives are used continuously where the data drives only when a backup is occurring.

it was 0.66% difference.

Now you could technically skew the numbers and say that "HDDS are twice as likely to fail as SSDs" but you'd still have basically the same amount of failures per 100 drives in either case.

 

[ScottSoapbox]

it was 0.66% difference.

Now you could technically skew the numbers and say that "HDDS are twice as likely to fail as SSDs" but you'd still have basically the same amount of failures per 100 drives in either case.

I'm not skewing the numbers, you simply aren't understanding them.

You can't compare failure rates (I.e., ratios) by subtracting them. Let me make it simpler for you: out of 10,000 drives you can have either 164 or 98 fail. Saying well only 66 fewer failed isn't a very meaningful comparison because it ignores the 10,000 part of the equation. So you do the math and see that SSDs failed 60% as often as HDDs or if you want to round about half as often.

And no, one can't say that "twice as likely" is "basically the same" just because you think the numbers are both small.

 

[Blqckout]

Idk what any of this means, but numbers are cool ig

 

[alexnode]

Seagate and WD! These are not great drives, the crucial MX series is not great too. I have about 30 SSDs from 2011 till now and I didn't have even one failure. Samsung and Crucial only. Some of them have seen 24/7 server use with a lot of writes.

 

[yRaz]

I'm not skewing the numbers, you simply aren't understanding them.

You can't compare failure rates (I.e., ratios) by subtracting them. Let me make it simpler for you: out of 10,000 drives you can have either 164 or 98 fail. Saying well only 66 fewer failed isn't a very meaningful comparison because it ignores the 10,000 part of the equation. So you do the math and see that SSDs failed 60% as often as HDDs or if you want to round about half as often.

And no, one can't say that "twice as likely" is "basically the same" just because you think the numbers are both small.

I mentioned 100 because I've owned probably 60-70 hard drives in my life and I've only ever had 1 drive failure. Statsticially, I can buy another 100 harddrives and not have a failure. We are getting into numbers that are meaningless to consumers. I don't care if HDDs are twice as likely to fail as an SSD if I have to buy more drives than my house is worth to see a statistical failure.

 

[BigMeatyClackers]

Many of the first ssds I purchased which used SLC and MLC memory cells are still going strong today. Other aspects of a drive can fail but I always got the impression that durability was undervalued by consumers.

 

[Superconductor]

I have various 'vintage' of Samsung 2.5" 850 & newer series SSD's and a few Samsung nvme "sticks" too and none of them have given me any problems. Doesn't matter if they were OEM grade or consumer, they all continue to perform as spec'd! ... some are older than 5 years & going strong.

 

[kiwigraeme]

Something to watch out - need more nuance.

ie there is drive failure and drive failure

eg can you recover data?
is there warning?
Is it good drive failure - ie in the first month - not saying this is good - but maybe not much on it - and if drive gets pass the first month - then long life ( applies to spinning rust as well )

probably what is more interesting is creatives who hammer their SSD and M2 drives for say endless video creation - often filling it to near capacity - SSD drives apparently don't like this .
Joe public running torrents day in day out - anecdotally I haven't seen stories about say samsung drives falling over (latest model story was a specific problem) .

But my main point know SSD drives can fail

 

[ddferrari]

I find it very hard to believe that a solid-state device, with no moving parts, is barely more reliable than an old-school HDD.

I call BS on whoever this "Backblaze" is.

 

[3ogdy]

My 128 GB SSD 830 (bought new) is still going strong, albeit it hasn't been used heavily for quite a few years since I upgraded the OS SSD

 

[kiwigraeme]

I find it very hard to believe that a solid-state device, with no moving parts, is barely more reliable than an old-school HDD.

I call BS on whoever this "Backblaze" is.

apparently controllers in them can be problematic - if they have - you are getting no data back - LED lights when they came out promised 10000 hours plus - the led part yes , but the voltage controller part ( or what ever its called ) maybe not

 

[KofeViR]

Seagate and WD! These are not great drives, the crucial MX series is not great too. I have about 30 SSDs from 2011 till now and I didn't have even one failure. Samsung and Crucial only. Some of them have seen 24/7 server use with a lot of writes.

I have had like 20+ Samsung, ~5 Crucial, ~10 WD and never had an issue either.

IMO Samsung lost momentum recently. WD does really well in terms of sales and my last 2 own SSD purchases were WD 1TB SN850 and then 2TB SN850X. I considered the 990 Pro but I was glad I didnt because of the issues with degradetion.

SN850X and 990 Pro is pretty much the best drives you can buy for consumer-use right now, so not great drives? Yes it is. WD Black line scores top scores in reviews and works flawlessly without firmware issues.

SN850X is best value for sure over 990 Pro and don't have issues with degradetion or firmware issues. However Samsung fixed them by know ("we think") I know a guy that had to pull the 990 Pro out from his PS5 in order to update firmware and put it back in... Would be nice if Samsung did not release beta products in the future, however I still like Samsung SSDs in general and have 4 running in PCs and servers.

 

[Avro Arrow]

That is very surprising to me, not because I thought that SSDs were overly-reliable but because I had thought that HDDs were more reliable than SSDs in the long-term. After all, SSDs have that write limit that HDDs don't. I suppose it doesn't matter much because I only use HDDs for backups and bulk media storage anyway.

 

[Hodor]

That's because it takes 5x longer for the system to boot from a HDD than SSD. So, you can only fit a certain number of reboots per day if you own a HDD. I mean before getting gray hair.

 

[Vanderlinde]

That is very surprising to me, not because I thought that SSDs were overly-reliable but because I had thought that HDDs were more reliable than SSDs in the long-term. After all, SSDs have that write limit that HDDs don't. I suppose it doesn't matter much because I only use HDDs for backups and bulk media storage anyway.

As drives become more complex, so was the reliability with it. It got worse over the years.

I have a few brands, samsung for example, 4x 320GB which is over 13 years old by now and still works perfectly fine. I used it as a backup, but 100% a seagate or wd woud'nt survive that timespan at all.

Ive done data recovery for some years for consumers or businesses, the majority of disks where always Seagate or WD's. Most common faillure was a corrupt firmware.

 

[rsmith222]

apparently controllers in them can be problematic

Sandforce 2 in OCZ Vertex 2 drives, especially the 240 GB version (the 64-bit NAND one that was actually sold to the public rather than the 32-bit one that was sent to reviewers).

Aside from bricking, Sandforce controllers weren't compatible with TRIM, despite claiming the contrary.

Sandforce hyped its controller according to its very low write amplification but that was worthless because the company chose to make it impossible to access one's data in a bricked drive — so people couldn't tamper with its precious firmware secrets. It also didn't create a robust protection system against the bricking in the first place.

Sandforce 3 also had a checkered reputation. Perhaps Intel's release of drives with that controller were a bit better.

Other notable examples that come to mind are the Intel G2, which was fixed with firmware, and the Samsung 840 series ('fixed' with firmware kludge). More recently, Samsung has been in the headlines again for problems.

 

[human7]

I find it very hard to believe that a solid-state device, with no moving parts, is barely more reliable than an old-school HDD.

I call BS on whoever this "Backblaze" is.

The headline is misleading. They found that SSDs had 40% fewer failure rates than HDDs, but since the failure rates for both are so small, the author of the article kept things in absolute terms and incorrectly claimed a slight difference (unless the author was quoting Blackblaze, in which case I want an explanation for why that relative difference isn't big enough to only make a claim for a slight difference).

This line in the article also concerns me: "An interesting tidbit about Backblaze's report is that the company hasn't used a single SSD unit made by Samsung, which is a major player in the SSD consumer market. One possible explanation is that Samsung drives aren't cheap, and Backblaze is essentially using the cheapest drives they can buy in bulk quantities." So, we're dealing with selection bias also.

It's also worth noting that of all the SSD drives, the ones that failed were all 250GB in capacity or smaller.

 

[human7]

That is very surprising to me, not because I thought that SSDs were overly-reliable but because I had thought that HDDs were more reliable than SSDs in the long-term. After all, SSDs have that write limit that HDDs don't. I suppose it doesn't matter much because I only use HDDs for backups and bulk media storage anyway.

That write limit might be coming into play here, though. All the SSDs that failed per the table were 250GB or lower in capacity.

 

[Pursuant4729]

It is worth noting the boot drives are used continuously where the data drives only when a backup is occurring.

The article states that BB started using SSDs as boot drives in 2018. What were they booting from before? I'm sure that the data analysis team have taken this into account and are using comparative data.

 

[zulu53]

I mentioned 100 because I've owned probably 60-70 hard drives in my life and I've only ever had 1 drive failure. Statsticially, I can buy another 100 harddrives and not have a failure. We are getting into numbers that are meaningless to consumers. I don't care if HDDs are twice as likely to fail as an SSD if I have to buy more drives than my house is worth to see a statistical failure.

True. What these numbers are saying to consumers is: SSD and HDD reliability is the same so if reliability is the criteria used for purchase just buy the least expensive per GB. If you want to evaluate based on access speed (likely the only criteria that can justify the increased cost of SSD) or energy consumption (again though: unlikely to be relevant to a NAS consumer) or space requirements (again though:unlikely to be relevant to a NAS consumer) then use these criteria and ignore a reliability criteria. Its your dollar, spend it wisely.

 

[Superconductor]

I find it very hard to believe that a solid-state device, with no moving parts, is barely more reliable than an old-school HDD.

I call BS on whoever this "Backblaze" is.

Their data is narrowed &d silo'd by products and not tech specific (categorically organized by nand type, etc...) so in many respects ... it's 'fun with statistics BS'. Don't even get me started about their "duty cycle" methodology.

 

[ChipBoundary]

All this illustrates is the point that there's literally no need for spinning drives anymore. We need to get rid of them and focus the production of those on SSD's to reduce price per/GB. SSD's have drastically higher max capacity right now, they are smaller, use dramatically less energy, use less physical materials, and are greener to produce.