Backblaze HDD Drive stats full 2023

[EAZ1964]

With the usual remarks that datacenter disk use differs from NAS use, and it is all statistics please find the link to the blog.

Backblaze Drive Stats for 2023

Read the 2023 Drive Stats Report and the latest insights on drive failure from Andy Klein. Read the 2023 Drive Stats Report and the latest insights on drive failure from Andy Klein.

www.backblaze.com

No real surprises (other than that 10TB disks had higher than average AFR, and 16TB lower).
Furthermore some nice new graphs and some less nice as I do not like the use of polynomes in the graphs because it proves that the underlying mechanism is not understood
have a good read.

 

[Coop777]

Can someone please explain to me what Backblaze "drive days" are? Is this a projected expectancy of an HDD lifespan?

 

[EAZ1964]

One drive spinning (operational) for one day is a one drive day. So a drive that has been operational for one year has 365 drive days and a drive that have been operational for one month has 30 drive days (or 28 or 31 depending on the month.) The two drives together have 395 drive days.

an important parameter as it defines the total time the drives have been working, many drive days give you better statistics.

 

[Coop777]

...which makes sense, but now I don't understand this:

with only 365 days in a year, how can any drive model have over 200,000 "drive days" ?

 

[EAZ1964]

because they have many disks! it is disks * days

#disks > 200000/365

 

[Coop777]

ah ok...so that's active disks "in the field" so to speak.

 

[EAZ1964]

yep, active disks * time run.
Many disks will increase the reliability. They take out the disk models with a less hours, to improve the confindence interval.

 

[Coop777]

This is the first BB report where I've seen an HGST Ultrastar with a 5.52% AFR. Highly unusual...although the driveline is ~ 9 years old.

 

[EAZ1964]

Please, don't look at statistics that way.
Picking an example with just 89 disks and 33k drive days is bad interpretation of the statistics.
It is like "my grandfather died at 105 and he smoked a package of cigarettes each day"
and conclude smoking is not bad for your health.

Data like this needs careful interpretation.

 

[Coop777]

Agree - it is how you slice and dice it, Ultrastars across the board remain rock solid.