F2FS Experimental Compression Is Ready For Extending Flash Storage Life

[skeevy420]

Ok I understand the explanation about the risks for the end-user. So the question is why to know about the amazing features of other file system if the end-user is prevented from their implementation. Every time I read articles of incredible improvements which are limited to few ones for many different reasons. As example, consider Wayland, an excellent evolution in terms of efficiency and reliability graphical stack. Only Fedora use Wayand by default since the last release, but Fedora has some limitation in fruition of some video codecs useful to see videos online, and or Tv contents. Other linux operating systems aren't able to integrate Wayland if not penalizing the usability.... So the question is, how can take benefit of all this improvements, if these improvements is not accessible to all the users? I think that is a logical matter much more than an informatics matter.

Because it takes time for those features to make it to the users. ZFS on root has been possible for many, many years, half a decade, but only for the more advanced and skilled Linux users who knew what they were doing and the limitations involved in doing so.

ZFS on Linux first came out in 2008. It wasn't until 2013 that it was usable. It wasn't until 2018 that Ubuntu started working on Zsys and ZFS integration. 20.04 is the first release planned for it general Ubuntu public use. Advanced users and sys-admins could have done most everything their tool does for the past few years, albeit not as simple as how Ubuntu it trying to make it.

Wayland also came out in 2008. It's just now getting to the point to where mainstream distributions are willing to risk using Wayland. Unlike ZFS above that needs wrapper tools to make it easier, everything Wayland has to be written from scratch and then implemented elsewhere.

Or it is one thing making Zstd. It's another thing adding Zstd to the kernel, QT archive programs, GTK archive programs, generic system archive libraries, game development engines, package management systems, and more while doing it all in a manner that all of Zstd's features are supported and working.

[evil_core]

So, how to get the feature to make longer the lifespan of an SSd, for end-user? The end-user is excluded from this features because ignore or is not able to run the way to integrate F2FS.

Typical users don't care about buying more RAM(and disabling SWAP) or to disable hibernation and use s2ram, so on typical 8-16GB RAM machine , he will wear his QLC flash very quickly dumping at least 40GB daily on SSD.
And typical user does so many stupid things, like running Avast Antivirus and other bloatware using much of I/O and CPU and RAM, so for performance it doesn't matter if he will ever use some experimental filesystem feature boosting SSD i/O by 2-20%/ They even don't are to set correct BIOS settings(or update it), or to even change from IDE emulation to native AHCI in the BIOS.
But most SSD killer is overfilling it and not using SSD Trim (while XP is dead and people swithing to at least window 7, native AHCI driver is needed for TRIM). And to lesser Flash wear you need to disable updating last access time attribute (you can dat using Nareon SSD Tools, even under Windows XP). Bigger problem is under Windows Tablets, where's eMMC, insead of proper SSD, and no trim support.
So as you see, there are more important problems with SSD, than some F2FS(if you have ever tried to use, it's much more slower nad problematic than BTRFS. I'm using it on on 2TB 960 Pro, and year ago it were unstable[twice filesystem became unmountable. Now it's stable, but mount times are long I wouldn't recommend it for average user. [XFS/EXT4/BTRFS are all massively tested on server workloads, much more optimized and works really nice on any FTL FLASH device])

[Ardje]

Use some distro with F2FS support on install, like Gentoo, Arch, Opensuse (I think)...

Any debian derived should be able to mkfs.f2fs by hand I guess. I use several filesystems to see what happens with it on usage. My steam lib is f2fs, my / is proven ext4.

[Ardje]

So as you see, there are more important problems with SSD, than some F2FS(if you have ever tried to use, it's much more slower nad problematic than BTRFS. I'm using it on on 2TB 960 Pro, and year ago it were unstable[twice filesystem became unmountable. Now it's stable, but mount times are long I wouldn't recommend it for average user. [XFS/EXT4/BTRFS are all massively tested on server workloads, much more optimized and works really nice on any FTL FLASH device])

YMMV. I have more positive results than I ever had with btrfs. Even though I retry it from time to time. I might retry XFS, because the last time I used that was 12 years ago, and it was nowhere near stable at that time. I have had f2fs as root filesystem on eMMC for quite a while without any problems.
To be honest: anybody using f2fs or btrfs should always bake a new kernel. New features and bug fixes. They target different audiences though. F2FS is for eMMC, microsd and cheaper SSD. btrfs is for SSD's with a very good FTL because it writes a lot more. And the cpu and memory load of btrfs is of course a lot bigger. So again for embedded, btrfs is not the right place, but f2fs might replace my ext4 installs.
I will try btrfs again on rbd and fingers crossed hope it doesn't take down my cluster ;-). I love the features of btrfs, but got hit by several bugs in production that usually meant starting all over losing TB's of data in one self corruption. While the losing part is not really the problem, it also takes a long time before a btrfs recovers after a self corruption. So when you decide to start all over, it's already been a week of lost production. But I think the biggest issue is: I was bitten by this when it was declared more or less stable, while it probably is only now stable.

[Ardje]

This could possibly increase performance on SSDs bottlenecked by slow bus (sata, usb...), am I right?

I wouldn't call SSD slow.
F2FS comes from Samsung, their main focus is mobile. They invested in wayland, v4l2-m2m for open video decoding/encoding which actually only applies to embedded systems, they invested in enlightenment as the major wayland compositor, they invested in gstreamer to get v4l2-m2m actually working and to have a chromium with gstreamer plugin for video with a wayland renderer.
Now mobiles usually come with eMMC's with write speeds of 45MB/s and read speeds of 100MB+ . LZ4 is not the fastest compressor, but it can easily fulfill that write speed target. On the decompression side, LZ4 outperforms anything else, and can have a real high speed.
Sata is not a slow bus compared to eMMC. And since everybody is switching to UFS (which is sata but with interchip voltages, and not sata bus voltages), you can see that even that's not the problem. The real problem is that tiny devices "do not have" fast SSD, because the "SSD" does not use that many flash chips, it actually is a single chip with flash and controller. Any current PC SSD on sata does wirespeed because they use a lot of seperate nand flash chips, a lot of dram and a smart controller connected to it and can do a lot of parallel writes. So on a PC I wouldn't use compression for that.

[Azrael5]

A question: is LZO/LZ4 able to optimize writing or the upgraded F2FS?

[Azrael5]

Another question: the current LZ4 is based on the MLZ4 provided of significant improvements in terms of compression and decompression abilities?