F2FS With Linux 4.4 Brings Better Stability/Performance For In-Memory Extent Caches

[phoronix]

Phoronix: F2FS With Linux 4.4 Brings Better Stability/Performance For In-Memory Extent Caches

Jaegeuk Kim sent in the Flash-Friendly File-System (F2FS) updates today for targeting the Linux 4.4 kernel...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-4.4-F2FS-Updates

[Azrael5]

I hope a better way to chose the file system since the beginning:a simple option where user can chose between ext4 f2fs and others according to its own mass storage.

First step: which file system? ext4 (HDD), or f2fs (SSD-flash memories) or btrfs (RAID)...

Second step: How many partitions!? 1 2 3 ... and 1 specific for pagefile.

One step after another step: the simpler the more user friendly, more users can be incentivated to familiarize with Linux systems.

[chris200x9]

How is F2FS even in the kernel? It's seriously unstable.

[dragonn]

How is F2FS even in the kernel? It's seriously unstable.

What do you mean? F2FS runs on my Arch ~half year on the system partition - no problems and it survived some dirty shutdowns too.

[chris200x9]

What do you mean? F2FS runs on my Arch ~half year on the system partition - no problems and it survived some dirty shutdowns too.

I was installing arch on f2fs and it crashed. Everything I had done within the last 10 minutes was gone, ie. adding a user and editing sudoers and stuff.

[caligula]

How is F2FS even in the kernel? It's seriously unstable.

How is btrfs? I tried it as a root file system with kernel 4.2 and already go checksum error in one month. No, the system doesn't have ECC. Disk is ok, no bad sectors.

[arter97]

How is F2FS even in the kernel? It's seriously unstable.

It's f2fs' characteristic, rather than stable-ness.
f2fs has proved itself to be stable from last year as many Android devices are mass produced with f2fs shipping by default.

[SystemCrasher]

First step: which file system? ext4 (HDD), or f2fs (SSD-flash memories) or btrfs (RAID)...

I guess most users just do not care. And those who do, could have their own idea on it anyway. Say, BTRFS isn't inherently bad for flash drives either and while it would be slower and less optimized for SSD, it would give you some goodies like compression, ablity to reflink, so you can "copy" big hierarchy in a blink of eye and without wasting space twice, advanced snapshots and so on. Or ext4 ... it would do okay on SSD, and it has been tested for a while, and bugs were more or less addressed. Because it is quite old and relatively simple. Btrfs is far more complicated, so it is kinda logical it contains more bugs and it takes more time to iron them out. And F2FS is relatvely new and not yet uber-popular, so it haven't received widespread testing.

Second step: How many partitions!? 1 2 3 ... and 1 specific for pagefile.

You can do something like this in *buntu installer: do not let it auto-partition things and choose to go on your own. Then graphic interface of partitioner appears and you can do exactly that: choose filesystems and partitions, whether to format them, which mountpoints they should use (swap included, etc). So it's already here for something like 10 years...

One step after another step: the simpler the more user friendly, more users can be incentivated to familiarize with Linux systems.

That's how I install xubuntu.. and you see, not everybody want to know how exactly filesystems are differ. And those who want to learn it can have their own idea how to do it right. Somehow, even *buntu installer is smart enough about this topic. It can also set up encryption of /home or even whole disk in relatively friendly way, hiding LUKS complexities behind fancy UI. Good example for others how to do it right. Though most desktop oriented distros wouls allow something similar.

[SystemCrasher]

How is btrfs? I tried it as a root file system with kernel 4.2 and already go checksum error in one month. No, the system doesn't have ECC. Disk is ok, no bad sectors.

Btrfs is fine, as long as you use >=4.0 kernel or so. I've been using it for more than year, and I can admit it is like this:
- It worked reasonably in something like 3.1x for me. Though it means I've been lucky, since there were some high-profile bugs and these were fixed later. But I only learned about bugs from commit logs, which is how I prefer to learn about bugs in filesystem
- Somewhere like 3.15 to 3.18 they were switching to kernel workers instead of homegrown implementation. Good thing in general, but since it was big intermal change, fallout followed. And it was like this: kernels (>=3.15) && (<3.19) would face noticeable memleak if you use compression in btrfs. Except this oddball it worked okay, though.
- Some bugs were fixed... and so far, fixes beyond 4.0 are not looking scary to me and more or less like any other fixes in any other filesystem. Maybe with exception of RAID5/6 code.

And no, I haven't seen checksum errors in whole year, in several configurations. Even after crashing kernel or losing power several times. So unless its something like (half-baked) RAID5/6, either you have stumbled on some new bug, or your hardware misbehaves and exposes rare errors. Chris Mason has already admitted they caught some few badass hard drive controllers silently damaging data at Facebook, thanks to checksumming abilities. In your shoes I would run heavy stbility tests on hardware and check for data corruption, unless you're in mood to face silent bit rot. If you have rare errors, and it goes undetected, eventually you can end up with either filesystem which is damaged to degree you can't mount it at all or some files can eventually turn out to be damaged. And its not a big fun when you need some file and then understand it rather contains some garbage. It can take several years to get there, but when it starting to get as bad as filesystem noticing damage via indirect ways, it could be way too late and half of filesystem could be already damaged at this point. Checksumming is good because it gives EARLY warning. And if you get one, it is something to consider. Checksumming part of BTRFS works okay on its own to the best of my knowledge.

[cynic]

How is btrfs? I tried it as a root file system with kernel 4.2 and already go checksum error in one month. No, the system doesn't have ECC. Disk is ok, no bad sectors.

Tell "how a filesystem is" is a quite hard challenge! There are literally tons of corner case in a filesystem code, and that's why they take decades to mature.

For what it's worth, here's my experience: I've been using BTRFS since before it was even merged in the official kernel repository: I have had a lot of troubles but since the latest releases it is running very smooth and reliably (both with single-device volume and multi-device).