Backblaze Hard Drive Stats for 2017

Glad we could help!

Aside from happily using Backblaze, their stats have convinced me to go with HGST for spinning rust several times now.

They aren't using many WD drives at all, and in the high use drives HSGT's roughly 1% failure rate against Seagate's roughly 2% failure rate is offset by price.

If the TCO / TB is lower, then it doesn't matter if they fail more often, it's already factored in.

As well as the risk/cost calculation already mentioned, it makes sense to spread over multiple manufacturers to allow for sudden problems with the one that is currently most reliable. If HGST has a huge production problem now that doesn't make its effects known until silly high rates of failure occur a year down the line, backblaze and other large storage providers would have a sudden glut of replacements to manage (possibly pro-active replacements, or even worse the expense of bulking up their redundancy level at least temporarily to mitigate the extra risk). Spreading over multiple manufacturers limits the impact of situations like this.

Yev from Backblaze here -> Mostly comes down to pricing! Since our service is designed for drive failure, we don't mind a bit of it here and there, but we are const-conscious, so if the drive failure rate < cost-per-GB, we tend to be OK w/ a bit more failure.

SSD is not a great long-term archival medium as it's not passively stable in the way that magnetised media is, and it's still much more expensive.

A spinning disk dissipates about 6W on average, for 100 000 disks that's 600kw. With an average of 0.12 USD paid for a kwh times 24 hours, their energy bill is less than 2000 USD a day. Their HDD cost based on the "81.76 — The number of hard drives that were installed each day in 2017." and we know they mostly now do 8TB at roughly 0.02 USD per GB according to https://www.backblaze.com/blog/hard-drive-cost-per-gigabyte/ which is 13000 USD a day. That means even if we had SSDs at zero watts, they would need to be (13+2/13) ~ 15% more expensive than HDDs in order to break even but if you look at https://pcpartpicker.com/trends/price/internal-hard-drive/ you will see it's still about 8 times as expensive. Obviously there's some savings due to floor space but the savings are going to be really low there. And this is just the economics side.

In addition to the other points, there are other issues besides just cramming a raw number of bytes into a raw amount of space.

You need to balance disk space with IO capacity with compute with memory.

Simply replacing all of the 3.5" drives with an equivalent volume of SSDs probably leaves you well under-balanced on the memory/compute side and way overbalanced on the IO side. In addition, you need to develop hardware to interface with way more drives per chassis (massive-scale expanders). There'd be a lot of engineering costs associated with getting that much denser in terms of the number of drives per U (figure 6 or 8 times as many per U).

Of course, this could pay off if there was a cost benefit to denser servers, but, datacenter space at their scale is primarily priced at a function of how much (and what quality) power you need. Being denser isn't ultimately going save them much if any money.

Not necessarily.

One nice thing about this report is they take into account the low statistics in some of their drive models and represent that as confidence intervals. If you look at the second table, 8TB HGST and 5TB Toshiba have a wide uncertainty due to them having tested fewer of these drives so far. Their data can not yet tell us if the AFR for these drives are significantly different from others.

They do have many of the smaller 4TB HGST and so have tight uncertainty on them and they are indeed low AFR. So if that size is suitable then it's a no brainer choice. But, if bigger capacities are needed and looking at the high confidence limit column, the 6TB and 8TB Seagate models are safer bets.

Yev from Backblaze -> Heh, clear cache and try again?

While you might have some legit grief with the service Backblaze provides, I would recommend toning the accusatory tone down by several notches. "questions sidestepped" "outrageous" etc. This is a courteous place.

This sounds like "Do you still hit your wife"-kind of questioning.

Yev from Backblaze here ->

1. Update: we're still trying to see if there's an elegant solution that would would be financially feasible. Since we try to run a sustainable business, it's tricky.

2. Update: We have a few internal proposals but thus far not on the engineering roadmap.

3. This one wasn't in the form of a question, but noted ;)