Hacker News new | ask | show | jobs
by pneumonic 1905 days ago
Showing the change in characteristics:

   One way to look at this is how long it would take to read the whole drive with a serial stream of 4kB random reads. The 1980s drive would take about 3 minutes. The SSD would take around 8 hours. The modern hard drive would take about 10 months. It's not a surprise to anybody that small random IOs are slow, but maybe not how slow. It's a problem that's getting exponentially worse.
2 comments

This (and write performance) is one of the reasons why mkfs now defers initializing block groups until the drive is mounted. After mount the kernel initializes them slowly in the background. I did a talk which covers this:

https://rwmj.wordpress.com/2019/02/04/video-take-your-loop-m... (right at the end of the talk)

Just the graphics: https://rwmj.wordpress.com/2018/11/04/nbd-graphical-viewer/

I was just explaining to a coworker that their RAID5 rebuild is likely going to over a day, during which time they shouldn't even think about any I/O intensive activities like snapshot merging in VMware.

The first time I did a RAID rebuild after a failed drive replacement was back around 2002 and it took just a couple of hours, if that.