Backblaze study on HDD reliability: Seagate HDDs are apparently garbage

This is a discouraging but welcome reminder to back up my stuff. I have a lot of important data on Seagate drives.

http://www.extremetech.com/extreme/1...

As you can see from the graph above, Hitachi drives are by far the most reliable. Even though most of Backblaze’s Hitachi drives are now older than two years, they only have an annual failure rate of around 1%. The “annual failure rate” is the chance of a drive dying within a 12-month period. After three years of being powered up 24/7, 96.9% of Hitachi drives are still running.

Western Digital is slightly worse, but still impressive: After three years of operation, 94.8% of Western Digital drives are still running. Backblaze lists the annual failure rate of the WD drives at around 3% (I don’t think the numbers quite add up, but I could be wrong).

Seagate drives are not very reliable at all. As you can see in the second graph below, Seagate drives are fine for the first year, but failures quickly start building up after 18 months. By the end of the third year, just 73.5% of Backblaze’s Seagate drives are still running. This equates to an annual failure rate of 8-9%.

This is the sort of thing that needs to be monitored on an ongoing basis as manufacturing and models change. Do they say if/when they will update?

Cripes! Glad I didn't pick up that 1tb Seagate HDD the other day.

As someone who works in the storage industry, I can say that even within a single manufacturer, there can be a lot of variation in reliability between models and even between production runs within the same model. Any company that buys drives in bulk (cloud providers, storage system manufacturers, etc.) is going to see spikes like that. If the drive manufacturer sees a spike in failure rates for a model, they're going to do failure analyses and fix the problem. You can't really tell ahead of time which currently manufactured drives are going to be crap. If you care about your data, back it up.

deftly wrote:

As someone who works in the storage industry, I can say that even within a single manufacturer, there can be a lot of variation in reliability between models and even between production runs within the same model. Any company that buys drives in bulk (cloud providers, storage system manufacturers, etc.) is going to see spikes like that. If the drive manufacturer sees a spike in failure rates for a model, they're going to do failure analyses and fix the problem. You can't really tell ahead of time which currently manufactured drives are going to be crap. If you care about your data, back it up.

I agree with this. I admin a data analysis site for the CMS experiment, currently we have 3.4 PB of disk on the floor. We get hot/cold results on almost all our drives. Yeah the 1.5 TB Seagates bit us in the butt but other Seagate models haven't been any worse that other vendor drives. Also the 1.5 TB Seagates were not supposed to be data center drives.

I wonder about confirmation bias on some of this stuff too.

Carl

If the drive manufacturer sees a spike in failure rates for a model, they're going to do failure analyses and fix the problem.

I haven't liked or trusted Seagate since the stiction problems, back in the late 80s. Hard drives were super expensive back then, and nearly all Seagate drives would fail at about a year, and they absolutely stonewalled on ever fixing or replacing any of them. And when I was selling PC clones, years later, we had more trouble with Seagates than with anything else. And when I was running a laptop fleet, we didn't actually lose very many drives, but the ones we did lose were most often Seagate. In that specific case, I wouldn't really call the Seagates "bad" in an absolute sense, but they were the worst of the ones that came in the laptops we were buying.

I am absolutely unsurprised that their current models aren't especially trustworthy.

Deftly, you probably work for a company that really cares about quality, and I suspect your belief that Seagate cares in the same way may be misplaced. Corporate culture doesn't change easily, and in my experience, Seagate has been among the sharpest of business operators, with all the negative connotations that implies.

I'd also been avoiding Hitachi drives, because of the "Deathstar" problems back when they were still IBM, but after seeing this article a month or two ago, I picked up a 4TB drive for myself, and it's been pretty good.

I don't do bulk hardware purchasing anymore, but I have quite a number of drives in my home network, and I'm starting to lose my Samsung 1TB drives after about seven years; two out of, um, I think six or seven, have died relatively recently, just in the last few months. I quite liked those drives, and I wish Samsung was still in the business, as I'd be happy to buy more. Seven years of continuous operation is just fine by me.

I also have a bank of eight 1TB WD RE2 drives, which have all been running for about six years now, and they're all still fine. One was complaining some when it was running in a hot room, to the point that the RAID controller offlined it, but once the AC was fixed, it's been fine ever since.

edit: those drives actually outlived the RAM on the motherboard they were originally connected to.

Really interesting perspectives, guys.

This certainly isn't the most scientific study out there, but I found it interesting just because of the large number of drives involved.

deftly wrote:

As someone who works in the storage industry, I can say that even within a single manufacturer, there can be a lot of variation in reliability between models and even between production runs within the same model. Any company that buys drives in bulk (cloud providers, storage system manufacturers, etc.) is going to see spikes like that. If the drive manufacturer sees a spike in failure rates for a model, they're going to do failure analyses and fix the problem. You can't really tell ahead of time which currently manufactured drives are going to be crap. If you care about your data, back it up.

Amen, brother.

I've seen drives from every manufacture fail.. as well as memory..motherboards..power supplies.. In general most of these worldwide probably even out statistically wise to roughly clump up. I've seen bad batches of just about everything.. Many years ago we ordered 3000 desktops from Compaq and got a bad run where out of 250 that Compucom staged for us 241 were bad. So I've seen some serious runs as well.. just recently even buying in a small batch resulted in an outlier I had 3 laptops I was taking to 32GB of ram (4 8's) and of the 12 I bought 6 were bad.. and this was from Crucial generally a reliable memory manufacture in my experience.

Sometimes companies will just create a model or line that has some design flaw as well which can skew the overall numbers.

I checked through my email archive, and it turns out the WD drives are RE3s, and they're about five years old, not six. And the Samsungs were accumulated over time, so I don't know how old they all are. One of the two that failed was manufactured 12/07, and the other was 2/08, so I got a little more than six years from them, not seven.

Looks I'm getting about a year ahead of myself.

Everyone's going to have anecdotal stories about which hard drives last longest. I have a Seagate terabyte HDD in my current machine that's been going since 2010. I bought a Western Digital to supplement it about a year later and it failed in a few months. *shrug* Like people have said, a lot of it can be one production run is simply better than others.

Hard to argue with the fact that one manufacturer fails more than the others, though.

Although, their bar graph is a little bit ridiculous.

Malor wrote:

Deftly, you probably work for a company that really cares about quality, and I suspect your belief that Seagate cares in the same way may be misplaced. Corporate culture doesn't change easily, and in my experience, Seagate has been among the sharpest of business operators, with all the negative connotations that implies.

I'd also been avoiding Hitachi drives, because of the "Deathstar" problems back when they were still IBM, but after seeing this article a month or two ago, I picked up a 4TB drive for myself, and it's been pretty good.

Nope, I'm under no illusions there. But if they're having to do a lot of warranty replacements (and I think all of these drives have a 1 year warranty at minimum), then it's usually in their own best interest to fix manufacturing problems.

I almost mentioned the Deathstar drives myself (I had two of them). They're actually changing owners again, Hitachi's 2.5" drive division is going to WD, and the 3.5" drives are going to Toshiba.

and I think all of these drives have a 1 year warranty at minimum

Sure, but if the drive fails at 1 year plus one day, it doesn't directly cost them anything. It might cause Internet outrage if the problem is bad enough, but everyone starts shouting 'anecdata!' unless a problem is truly dire. (e.g., Deathstars.)

The thing is, we get that quote wrong: the original quote is that the plural of anecdote is data. With a large enough sample size, you can draw conclusions, and Backblaze has a large sample size.

However, I was thinking about this some, and their sample is not quite saying what we immediately assume. If you read the original Backblaze blog entry, what they're sampling with the Seagate and Hitachi drives are the cheapest drives available at a given size. And they're comparing those with older midpriced WDs. They stopped buying Western Digital because they didn't make drives that cheap.

So, we can say with a fair bit of certainty that cheap Seagate drives, across multiple generations and densities and years, aren't all that great. We can also say with a fair bit of certainty that cheap Hitachi drives are very reliable. But we can't say that WD is better, because we're not testing the same thing there: the sample size is smaller, and the drives cost more. It's quite possible that more expensive Seagates might be just as good as the WDs. They even mention that the Barracuda LPs were well-made.

So I think the only true takeaway here is that even cheap Hitachi drives are very solid. It's a shame they didn't change the Deskstar name, because that's the biggest reason I haven't tried them for so long.

You have to wonder about any data coming from a company that subjects low end consumer drives into environments that those drives were never intended to perform in. Then taking that data and saying it will apply to normal consumer environments is suspect... We are even seeing drives from the major manufacturers that are focusing on consumer trends like NAS devices which should tell you something that these guys are broadening their product lines based on changing consumer and SMB trends.

Granted I'm not in the cheap online storage business and with proper data protection and redundancy strategies I can see why this company would take this approach but in my world and in the companies I worked in the past I would have looked like a fool sticking consumer HD's into my SAN's. I see that Google released a HD failure study some years back as well and they take a similar approach (consumer grade HD's)

http://static.googleusercontent.com/...

(Quick Highlight.. as temp's rise so do the failure rates.. the no duh moment but many people never think about their HD's temps since chances are most of us don't pack HD's that densely anymore.. but even my 16TB Drobo has had much higher failure rates than any drive I've ever put in a PC case)

So don't think I'm knocking BackBlaze's strategy.. I get it and it makes perfect sense.. I'm just cautioning that extrapolating the same results into real world consumer results is probably not accurate.. But I would certainly lean towards buying those models they list that have the lowest failure rates since after all if a drive can survive in "torturous" conditions then chances are it will survive even longer in my tower cases.

I read Backblaze's blog post a month or so ago. If the ET piece is just a precis, be sure to read the original. BB are very up front about the small WD sample, apples vs oranges, etc.

As an added anecdote, the company I work for has been ... well, we've had enough RMAs of Seagate drives over the last year that it has changed purchasing decisions for our data centers.

Yeah, I always buy Western Digital drives, even if they are more expensive. I have had issues with Seagate drives so often it felt like more than random chance to me.

We got the 1.5 TB Seagates with a 5 year warranty. Joke was on them!

(Not really, we were very unhappy.)

Carl

carljetter wrote:

We got the 1.5 TB Seagates with a 5 year warranty. Joke was on them!

(Not really, we were very unhappy.)

Malor wrote:

I haven't liked or trusted Seagate since the stiction problems, back in the late 80s.

That's funny, I used to have a bias against Western Digital because of stiction issues on their 1 GB drives back in the mid-90s. Back when I worked at a little mom and pop computer shop I was regularly having to tap on these like a stuck alternator so I could recover data.

I'd also been avoiding Hitachi drives, because of the "Deathstar" problems back when they were still IBM

Oh man, somehow I forgot about this. Probably some sort of self-imposed amnesia to block out the trauma. I had bought a lot of these before we started seeing failures. It got to the point where we just ate them because we were RMA'ing the replacement drives even. I had a huge stack of them at one point. I guess I should've kept my Never Forget obelisk.

The one thing I've learned in the 30ish years I've been working with computers is that every brand has their periods of difficulty. Every single hard drive brand (as just about everything else) has had a few years when their stuff wasn't great. The quality dips, then it tends to come back up again. Quality is definitely cyclical in this industry.

That's funny, I used to have a bias against Western Digital because of stiction issues on their 1 GB drives back in the mid-90s.

Well, the biggest reason I was pissed at Seagate was because the drives almost always failed just out of the 1-year warranty, and they absolutely would not fix any of them, or even admit there was a problem. Like I said, very sharp business operators: they give you exactly what they promise, and not one bloody iota more. And this was back when hard drives cost like a thousand dollars, and in an era when a thousand dollars meant quite a bit more than it does now. It'd be more like buying a $2500 piece of equipment today, with almost every one failing at just a little more than a year.

I don't remember the WD problem. Did they at least fix the bum drives?

Quality is definitely cyclical in this industry.

True, but Seagate seems to cycle consistently lower than just about anyone else.

LiquidMantis wrote:

I guess I should've kept my Never Forget obelisk.

Star Trek Voyager reference?

Malor wrote:

True, but Seagate seems to cycle consistently lower than just about anyone else.

If I personally had to score someone as having the historically lowest reliability, it would without a doubt be Maxtor. But then Seagate bought Maxtor so...yeah. I miss Quantum.

Personally, I run a Western Digital Black as my main HDD (with a Samsung SSD) and I have two Seagate 3TB drives where my media, YouTube projects and other archives live. Mostly because the Seagate 3TB drives are cheaper than equivalent Western Digital. But the majority of that stuff is also backed up to CrashPlan which is why I didn't care that much.

Yeah, my personal worst experience with hard drives was with Maxtor as well. They shipped out a bunch of drives that would eventually come down with a problem called high fly write. The aerodynamic properties of the heads was changing (I think from spindle lubricant), causing them to lift too far off the platter.