F2FS Prepares Some Minor Improvements For Linux 6.5

[phoronix]

Phoronix: F2FS Prepares Some Minor Improvements For Linux 6.5

The Flash-Friendly File-System (F2FS) is tacking on a few new features for the Linux 6.5 kernel...

https://www.phoronix.com/news/F2FS-Linux-6.5

[wallentx]

I'll check out the contents of this patch and see if it resolves my issue

https://github.com/jaegeuk/f2fs-stable/issues/9

[carewolf]

So with everything being SSDs aka flash drives now, what is stopping it from being a default FS? Did they ever solve the performance problems with SQL https://www.phoronix.com/review/clear-linux-f2fs ?

[Joe2021]

So with everything being SSDs aka flash drives now, what is stopping it from being a default FS?

AFAIK, you should not confuse SSDs and Flash. SSDs do have their own wear leveling and other abstraction layers above their flash chips. F2FS targets for "raw flash" drives without own intelligence and protection. The OS/FS has to take care for the well being of the drive then, but not with a modern SSD.

[Weasel]

AFAIK, you should not confuse SSDs and Flash. SSDs do have their own wear leveling and other abstraction layers above their flash chips. F2FS targets for "raw flash" drives without own intelligence and protection. The OS/FS has to take care for the well being of the drive then, but not with a modern SSD.

It doesn't do harm on SSDs though.

SSD firmware is not magic. Wear leveling still adds some write amplification.

[Joe2021]

It doesn't do harm on SSDs though.

SSD firmware is not magic. Wear leveling still adds some write amplification.

But it is pointless. F2FS tries to avoid to write to the same sector number too often, to avoid hammering this sector. This is very important for "dumb" devices. But a SSD uses virtual sector numbers with a volatile mapping to physical sectors. It is just pointless to try to avoid to not write too often to a specific virtual sector, when the SSD is mapping this virtual address to different physical sectors over time anyways.

It is true that it does not do harm. But when F2FSs key selling point is rendered obsolete, why not use a FS with features you actually need?

[Democrab]

AFAIK, you should not confuse SSDs and Flash. SSDs do have their own wear leveling and other abstraction layers above their flash chips. F2FS targets for "raw flash" drives without own intelligence and protection. The OS/FS has to take care for the well being of the drive then, but not with a modern SSD.

You're thinking of the older flash-orientated filesystems such as YAFFS and NILFS which take care of the wear levelling, etc on their side of things because they're assuming it's just "dumb flash" but F2FS is specifically designed for NAND with an FTL equipped and the dev mentioned SSDs as a use-case. I use it for my boot drive because it's consistently showed some of the faster application start-up times whenever I've seen it benchmarked, supports the features I need for my boot drive and has proven more than reliable enough so far, but tend to stick to ext4/xfs/btrfs for my bulk/main storage depending on what the usecase is and what features I want.

Here's the LWN article from when F2FS was first merged upstream where the dev outright says it's aimed for NAND with an FTL and labels SSDs as a use-case for it.

[Joe2021]

You're thinking of the older flash-orientated filesystems such as YAFFS and NILFS which take care of the wear levelling, etc on their side of things because they're assuming it's just "dumb flash" but F2FS is specifically designed for NAND with an FTL equipped and the dev mentioned SSDs as a use-case.

Well, maybe I have confused F2FS with other flash-oriented FS. Will do some research. Thanks for that info.

[Weasel]

But it is pointless. F2FS tries to avoid to write to the same sector number too often, to avoid hammering this sector. This is very important for "dumb" devices. But a SSD uses virtual sector numbers with a volatile mapping to physical sectors. It is just pointless to try to avoid to not write too often to a specific virtual sector, when the SSD is mapping this virtual address to different physical sectors over time anyways.

Yeah, but this causes write amplification. F2FS's metadata is structured like a log-structured filesystem, and those can work even write-once just in append mode. Of course, at some point the chain becomes too long, that's why there's a garbage collector built in.