The hosts were going through Backblaze's latest hard-drive failure report, the dataset most people reach for when they want to know which drives to buy. Salter's argument is that the numbers describe Backblaze's purchasing strategy more than they describe drive quality: the company buys cheap because failure is expected and its infrastructure absorbs it. He finishes the thought by saying that if you aren't willing to do serious analysis on those numbers, you probably shouldn't be reading them at all.
I'm going to have to look now. I don't think I want to, though.
There is that. You should also maybe read a little bit into some of the numbers of which drives Backblaze bought. For example, while the Toshibas had that slightly higher failure rate, they kept buying them. I wonder why. Is it because maybe they were less expensive, even though they had a slightly higher failure rate? Or were they just more available? Or, you know, there was some reason why Backblaze kept buying them anyway. And, you know, it doesn't mean that the Toshiba drives are bad.
If there's one thing reading something like a decade's worth of Backblaze's failure reports has taught me, it's that Backblaze buys as much of whatever they can that's as cheap as possible. And that's pretty much the criteria. To be fair, you know, they are a bulk storage company, and it absolutely makes sense for them to look at hard drives as something that not might fail and we'd like to prevent that, but will fail. And we just need to design policies that catch those failures before they become larger problems. So what makes sense for Backblaze may not make sense for you. And if you aren't willing to do some serious crunching and analysis when you're looking at Backblaze's numbers, maybe you shouldn't be looking at Backblaze's numbers because they frequently don't mean what you think they do.
Yeah, if you have the infrastructure to keep replacing the drives when they fail, a drive that fails more but costs less is an easy trade-off to make. If each time you have to buy a drive or deal with a warranty, it's an issue for you, then you might be willing to pay a little bit more for drives that don't have that.
And I think it makes sense that we're seeing much lower failure rates in the much larger drives, not only because they're newer, so they haven't had a chance to get to the crappy back side of the bathtub curve, but also keeping in mind, again, that Backblaze buys cheap drives. They want to get the cheapest thing that they can because failure is not only acceptable, it's expected for them. So they're looking for a good deal. And when you look at smaller drives, there used to be cheap drives and expensive drives at pretty much every size point, up until you got to eight terabyte drives. That's really when things kind of started to diverge a little bit. I would argue that when you look at like 16 terabyte or 20 terabyte drives, there really aren't any cheap ones anymore because it's not something that might wind up just being the C drive in somebody's budget laptop or desktop. If you're looking to buy a 16 or 20 terabyte drive, that's specialty equipment. And so I think there's a much higher bar to entry in terms of making something that doesn't suck if you expect to sell a 20 terabyte drive for what that's necessarily going to cost.
Changing language for storage capacity. This is a short piece from Michael Lucas, who both of you know.
