Backblaze Drive Stats for Q1 2024

The endurance figures for hard drives are probably derived from the rated number of seek operations for the heads, which is why it doesn't matter whether the operations are for reading or writing data. But that bakes in some assumptions about the mix of random vs sequential IO. And of course the figures are subject to de-rating when the company doesn't want the warranty to cover anything close to the real expected lifespan, especially for products further down the lineup.

Where do you buy your drives? Last time I was in the market, I couldn't find a reputable seller selling the exact models in the report. I'm afraid that the less reputable sellers (random 3rd party sellers on Amazon) are selling refurbished drives.

I ended up buying a similar sounding but not same model from CDW.

I don’t buy hard drives based on these reports. I buy SSDs and let my cloud providers deal with hard drives.

> That's a milestone. Imagine the racks that were eliminated

I'm imagining about 3/4ths ;)

Perhaps not eliminated, but repurposed with fresh 16TB drives. And the power savings per byte stored!

Yeah, but just thinking about it reminds me how annoyed I am that they increased the B2 pricing by 20% last year.

Right after launching B2, in late 2015, they made their post about storage pod 5.0, saying it "enabled" B2 at the $5/TB price, at 44 cents per gigabyte and a raw 45TB per rack unit.

In late 2022 they posted about supermicro servers costing 20 cents per gigabyte and fitting a raw 240TB per rack unit.

So as they migrate or get new data, that's 1/5 as many servers to manage, costing about half as much per TB.

It's hard to figure out how the profit margin wasn't much better, despite the various prices increases they surely had to deal with.

The free egress based on data stored was nice, but the change still stings.

Maybe I'm overlooking something but I'm not sure what it would be.

In contrast the price increases they've had for their unlimited backup product have always felt fine to me. Personal data keeps growing, and hard drive prices haven't been dropping fast. Easy enough. But B2 has always been per byte.

And don't think I'm being unfair and only blaming them because they release a lot of information. I saw hard drives go from 4TB to 16TB myself, and I would have done a similar analysis even if they were secretive.

If you are a large company owning hundreds of thousands of them and knowing you will have disk failures regardless, maybe. If you own just a few hundreds and a failure costs you money the logic may be completely different.

I'd assume so. Also consider that if a drive fails under warranty, and you're already dealing with a bunch of failing drives on a regular basis, the marginal cost to get a warranty replacement is close to zero.

How would they lead you astray? I wouldn't consider a drive failure in a home NAS to indicate that - even their most statistically reliable drives still require redundancy/backup - if you haven't experienced a drive failure yet, that's just chance.

Which specific ones do you like so far?

Backblaze has been slowly releasing SSD stats, but their usage of HDDs dwarfs SSD usage, so it won’t be as useful.

https://www.backblaze.com/blog/ssd-edition-2023-mid-year-dri...

I can't think of any reason why the lifetime would be any different for a refurb. of course, you need to start from when the drive was originally used. of course, there is probably also some additional wear and tear just due to the removal, handling, and additional shipping of the drives.

In some ways that would be incredibly noisy to test. However it could be a good way to measure the practicality of S.M.A.R.T metrics. Finding out how accurate they are at predicting hdd lifespan would be a great finding.

Refurbed drives, at least for Seagate, are terrible.

> Are most drives retired without failing?

I'd expect so given that HDDs are still having significant density advancements. After a while old drives aren't worth the power and sled/rack space that could be used for a higher capacity drive. And, yeah, it makes these statistics make more sense together.

Edit: plus they are just increasing drive count so most drives haven't hit the time when they would fail or be retired...

"Are most drives retired without failing?"

Yes, certainly.

One can watch both SMART indicators as well as certain ZFS stats and catch a problem drive before it actually fails.

I like to remove drives from zpools early because there is a common intermediate state they can fall into where they have not failed out but dramatically impact ZFS performance as they timeout/retry certain operations thousands and thousands of times.

Yes, and because of that the numbers on the average time to failure are completely meaningless. The drives the don't ever fail skew the numbers completely. If a fantastically reliable drive were to have 5/5000 drives fail, but they all failed in the first month and then the rest carried on forever, then that would show here as having a lower "reliability" than a dire drive where 4000/5000 drives fail after a year.

I'd like to see instead something like mean time until 2% of the drives fail. That'd actually be comparable between drives. And yes, it would also mean that some drive types haven't reached 2% failure yet, so they'd be shown as ">X months".

This is what a Kaplan-Meier survival curve was meant for [0]. Please use it.

Also, it'd be great to see the confidence intervals on the annualised failure rates.

[0] https://en.wikipedia.org/wiki/Kaplan%E2%80%93Meier_estimator

Obviously yes. At an AFR of 1.5% they'd have to have the drives run for (about) 67 years to have them all retire from failure.

(in reality they'd probably have failure rates spike at some point, but the idea stands. And they explicitly said they retired a bunch of 4TBs)

Drives have warranties, after which the manufacturer doesn't make any claims about it's durability. This could put your fleet at wild and significant risk if things start hitting a wall and failing en masse. You may not be able to repair away if as you're repairing the data you're copying to yet another dying drive.

So you have usually a lifetime of drive tput and start/stop values you want to stay under, and depending on how accurate your data is for each drive you may push beyond the drive warranties. But you will generally stop before the drive actually fails.

They just retired 4TB ones.

While they seem to get retired, it's not as quick as we'd think.

This uncertainty should be accounted for in the confidence intervals of their stats.

The WDC models which are only somewhat more expensive than Toshiba or Seagate tend to perform quite a lot worse than those. Models with the same performance are significantly more expensive.

Anecdata is such a weird thing. In my own NAS, I've had 3 out of 3 WD Red drives, each a different size, die in an identical manner well before their warranty expired over the last several years. SMART says everything is fine, but the drive's utilization creeps up to a constant 100% and its write IOPS decrease until the whole array is slow as frozen molasses. That's in a constantly comfortable operating environment that's never too hot, cold, or otherwise challenging. And yet it looks like I'm the statistical outlier. Other people -- like Backblaze here -- have decent luck with the same drives that have a 100% failure rate here.

Probability is a strange thing, yo. The odds of a specific person winning the lottery are effectively 0, but someone's going to. Looks like I've won the "WD means Waiting Death" sweepstakes.

I've had so many Seagate drives fail that I won't buy Seagate again.

If a brand sells bad drives, they should be aware of the reputational damage it causes. Otherwise there is no downside to selling bad drives.