Five years of data show that SSDs are more reliable than HDDs over the long haul(arstechnica.com)
arstechnica.com
Five years of data show that SSDs are more reliable than HDDs over the long haul
https://arstechnica.com/gadgets/2022/09/five-years-of-data-show-that-ssds-are-more-reliable-than-hdds-over-the-long-haul/
8 comments
I've run the past fileservers for about 10 years before rotating the data set to a new machine with roughly twice the usable capacity.
The price of rust is so much cheaper that I doubt I'd get anywhere close to ROI with SSDs, and I'd not feel comfortable running them for much more than 10 years anyway.. Even when factoring in power usage.
> I doubt I'd get anywhere close to ROI with SSDs
I think most people assume the SSDs are more expensive, but depending on their application they choose SSDs for the OTHER properties, like SSDs are quieter, and faster, and like you mention use less power.
There was this narrative for a long time that SSDs failed way more often, or beating them up with reads and writes would make them fail while hard drives could take the same number of reads and writes and not fail. So a subset of people chose hard drives for durability regardless of cost. Personally I never saw any evidence of SSDs failing more often than hard drives, and PLENTY of counter evidence. For instance, I wrote a program that unintentionally overwrote the same fixed length 16 byte file in the same location over and over again too often, and it failed more on drives than SSDs. (The failure mode was not whole disk failure, the 16 byte file itself mysteriously stopped being writable on some computers after a long time of working.) I had some friends that worked at Tivo and they said they had the same bug/shortcoming early on. The Tivo would overwrite the same section of disk over and over and it failed LONG before the drive manufacturers said it would fail. The drive manufacturer said they needed to vary the location of where they wrote the data to get anywhere near the published numbers.
I had assumed the file system layer would take care of that kind of management for me, but I guess not so much.
I think most people assume the SSDs are more expensive, but depending on their application they choose SSDs for the OTHER properties, like SSDs are quieter, and faster, and like you mention use less power.
There was this narrative for a long time that SSDs failed way more often, or beating them up with reads and writes would make them fail while hard drives could take the same number of reads and writes and not fail. So a subset of people chose hard drives for durability regardless of cost. Personally I never saw any evidence of SSDs failing more often than hard drives, and PLENTY of counter evidence. For instance, I wrote a program that unintentionally overwrote the same fixed length 16 byte file in the same location over and over again too often, and it failed more on drives than SSDs. (The failure mode was not whole disk failure, the 16 byte file itself mysteriously stopped being writable on some computers after a long time of working.) I had some friends that worked at Tivo and they said they had the same bug/shortcoming early on. The Tivo would overwrite the same section of disk over and over and it failed LONG before the drive manufacturers said it would fail. The drive manufacturer said they needed to vary the location of where they wrote the data to get anywhere near the published numbers.
I had assumed the file system layer would take care of that kind of management for me, but I guess not so much.
> the same fixed length 16 byte file in the same location
On the SSDs you weren't writing to the same location. The firmware keeps mapping your writes to different locations for wear leveling. So if those 16 bytes are the only thing you're writing your SSD would last a very long time indeed!
> I had assumed the file system layer would take care of that kind of management for me
Depends on the file system; some do, some don't, and for some uses cases that isn't what you would want, so you need to chose your filesystem (or its options) appropriately. A journaling file system (like ext4 with -j) would never write to the same spot (but without -j it would), and a more advanced one with COW like btrfs usually won't either (but COW can be turned off).
On the SSDs you weren't writing to the same location. The firmware keeps mapping your writes to different locations for wear leveling. So if those 16 bytes are the only thing you're writing your SSD would last a very long time indeed!
> I had assumed the file system layer would take care of that kind of management for me
Depends on the file system; some do, some don't, and for some uses cases that isn't what you would want, so you need to chose your filesystem (or its options) appropriately. A journaling file system (like ext4 with -j) would never write to the same spot (but without -j it would), and a more advanced one with COW like btrfs usually won't either (but COW can be turned off).
"as a boot device"!
I don't see that phrase in the article
"Backblaze uses SSDs as boot drives for its servers rather than data storage, and its data compares these drives to HDDs that were also being used as boot drives."
But a 13 year old commenter on a l33t gam3r forum told me I should never never use an SSD!
I still think digital storage in the modern era is a bag of hurt.