The Backblaze stats are nothing short of amazing, transparency-wise, and if anyone on the team making these available is reading this, many thanks to you!<p>That being said, truly bad series of HDDs have been extremely uncommon in the past decade. Of course, everyone is on NVMe Flash now for storage that actually <i>matters</i>, but I think the last epic "enterprise scale" nearline-class mess-up were the Seagate Barracude 1(.5)TB drives, introduced in 2013 and last sold in 2016?<p>Anyway, I've been very happy recently with some huge arrays running 14/15/18TB Western Digital SAS drives. The only (but pretty foreseeable) issue is that RAID6 arrays, once a drive in the set goes bad, simply can't be rebuilt in an acceptable timeframe.<p>So, standard procedure now is to have one empty RAID6-backed volume on standby at all times, so data can be migrated there from a degraded volume, after which the latter (after drive replacement) can be rebuilt from scratch.<p>Not an approach that would work at huge scale, but for simple-old-me, it's sufficient...
The blog post that this image is cherry picked from ("Backblaze Drive Stats for Q3 2023") was posted 17 days ago[0] (275 points, 72 comments), not sure why this needs a duplicate post.
[0]: <a href="https://news.ycombinator.com/item?id=38263435">https://news.ycombinator.com/item?id=38263435</a>