Samsung Introduces New Linux File-System: F2FS

[Ex-Cyber]

A very common SSD (Crucial M4) has surpassed 750 TB host write. Spread it over 10 years (3650 Days) and you'd have
to write over 200 GB (the drive itself will have either 128 or 256 GB capacity) per day every day for 10 years and it will still function.
SSDs can take a lot of shit contrary to popular belief.

What gets degraded in flash memory is the data retention time. It will "still function" in some sense almost indefinitely; the question is how long you expect it to hold on to your data. Typical specs I've seen are 10 to 25 years retention after 10,000 to 1,000,000 write cycles, depending on the technology. Of course you can push it beyond that number of writes if your data only has to persist for, say, 3 months. Rumor has it that the newest chips actually have worse numbers for this due to the smaller cell size (e.g. 10 years at 3000 write cycles), but all those datasheets seem to be under NDA so I can't verify this.

[Adarion]

Humm, I hope Samsung is not reinventing the wheel here.
I mean, I like file systems but we do have an awful lot of them in the kernel, for every purpose at least two. And we should make sure these FS are kept in good state instead of adding even more that are just halfway finished.
Furthermore there are supposed to be a couple of FS that deal with the issue of writing multiple times on flash memory, and as mentioned by people here, flash became more robust in the last 5 years than it was before.

I mean, couldn't these Samsung guys have participated in one of the already existing projects?

[oliver]

Samsung is the world leader in flash technology.

Why would they spend money on this if they didn't think it was worthwhile?

Well, they did this once, with 'RFS' [1] (and that's still in the wild lively today). Robust Filesystem .... a joke!
http://lwn.net/Articles/362020/
That said, it could be a proper attempt to rectify their previous attempt

So while I applaud their efforts and I find Samsung an awesome company (for some of the things they do I suppose) if this falls into the same category ... hah.

[zeealpal]

You do realize that NAND is a type of flash memory, right? (There's also NOR flash, which is commonly used when direct addressing is required, e.g. BIOS ROMs.)

I think he does. I think he is also pointing out that the quality and controller differences, among other things differ greatly, hence why SSD memory != USB Flash memory in read/write counts before failure.

[TAXI]

Unlike lanyardfs, this one actually has a point. It'll maximize the life of the flash.

btrfs will not do that. Any generic fs will not do that.

Linux really has been lacking a filesystem tuned to the specifics of NAND flash, and this fills the gap.

http://en.wikipedia.org/wiki/YAFFS
http://en.wikipedia.org/wiki/JFFS2
http://en.wikipedia.org/wiki/LogFS
http://en.wikipedia.org/wiki/UBIFS

I never heard of anyone running into problems because his router wrote to much data to often to its flash storage.

Maybe because that router firmwares you're referring to use JFFS2 (or squashfs + ram discs) ?

[dalingrin]

I always see a ton of people on this forum complaining about what other people choose to work on. I find this attitude pretty odd. How is another filesystem going to hurt you? Filesystems are very modular, they require very very little maintenance to keep them usable in kernel updates. If a company, that is responsible for a large part of the flash memory out on the market, wants to put their resources into making a filesystem that is tailored for their market then they should be able to do it. My guess is that they know a little more about their own flash memory than you do.
God forbid, not only did they do this work but they also wanted it to be open source...yes, the best course of action is to complain.

[Adarion]

Nah, we're not complaining about people doing free software. But it is a bit sad to see such a lot of fragmentation in filesystems. I think it would sometimes be better to have a concerted, focused effort and bring some thing to 100% and shinyness than to have split forces and everybody just crawling forward at 33% completion.
But if their new FS really brings totally new points, esp. in handling flash memory, well then I will accept this new FS gratefully. (Though there are efforts for 5 or more years now to write various FSs that actually focus on handling flash media.)