ATGC Could Come In Linux 5.10 For F2FS, Much Faster Decompression Speeds Too

[phoronix]

Phoronix: ATGC Could Come In Linux 5.10 For F2FS, Much Faster Decompression Speeds Too

We previously reported on F2FS "ATGC" functionality for increasing the garbage collection efficiency for the Flash-Friendly File-System. Those patches are now queued up in F2FS' "dev" branch meaning we could see the functionality in place for Linux 5.10...

http://www.phoronix.com/scan.php?page=news_item&px=F2FS-ATGC-Dev-Branch

[tildearrow]

On the Android Pixel 3

That's a bit of an odd name for the Google Pixel 3...

[uid313]

Is F2FS the future of file systems?

Soon SATA, SCSI, and SAS are no more in favor of M.2 and PCI Express. Mechanical HDDs are fading away in favor of SSDs which are growing in size, now there are SSD in the range of terabytes. Cylinder-head-sector (CHS) is going away.

[skeevy420]

Is F2FS the future of file systems?

Soon SATA, SCSI, and SAS are no more in favor of M.2 and PCI Express. Mechanical HDDs are fading away in favor of SSDs which are growing in size, now there are SSD in the range of terabytes. Cylinder-head-sector (CHS) is going away.

IMHO, those reasons are why file systems like ZFS and BTRFS are growing in popularity among the end-user, non-enterprise, desktop using crowd. Aside from price per TB, the biggest benefit that mechanical HDDs still have is longevity when compared to SSDs which is why a robust file system that can do RAID and, therefore, can be setup to handle disk failures w/o data loss is more necessary in the day and age of SSDs that don't necessarily live as long as HDDs.

But for single disk scenarios F2FS is looks like a better and better choice every time Phoronix has an article on it.

[uid313]

IMHO, those reasons are why file systems like ZFS and BTRFS are growing in popularity among the end-user, non-enterprise, desktop using crowd. Aside from price per TB, the biggest benefit that mechanical HDDs still have is longevity when compared to SSDs which is why a robust file system that can do RAID and, therefore, can be setup to handle disk failures w/o data loss is more necessary in the day and age of SSDs that don't necessarily live as long as HDDs.

But for single disk scenarios F2FS is looks like a better and better choice every time Phoronix has an article on it.

Well, while ZFS is an innovative file system I think it is not aimed primarily at flash storage and is very well suited to mechanical storage. The same goes for Btrfs.
So perhaps ZFS and Btrfs can be described as "innovative old school file systems".

F2FS on the other hand is designed explicitly for flash storage, which is rather interesting, but I don't know how good it is. Also, I don't know if it is any useful since most flash storage devices have a Flash Translation Layer (FTL) which emulates a traditional storage device.

Most file systems and disk I/O schedulers were designed for mechanical hard drives built on cylinders and heads on top of legacy protocol such as SATA, SCSI and then SAS which is a modernization of old SCSI.

[tildearrow]

Is F2FS the future of file systems?

Soon SATA, SCSI, and SAS are no more in favor of M.2 and PCI Express. Mechanical HDDs are fading away in favor of SSDs which are growing in size, now there are SSD in the range of terabytes. Cylinder-head-sector (CHS) is going away.

This is why I compete with you.

That's too radical for enterprises and users in developing countries.
No, thanks.

[zyxxel]

Is F2FS the future of file systems?

Soon SATA, SCSI, and SAS are no more in favor of M.2 and PCI Express. Mechanical HDDs are fading away in favor of SSDs which are growing in size, now there are SSD in the range of terabytes. Cylinder-head-sector (CHS) is going away.

Well if I need a couple of hundred TB of storage space then M.2 isn't that easy to use. With SAS/SATA I can use 25 20TB disks - if I manage 6 SSD, then each of these SSD needs to be awfully large... There aren't that many 80 TB SSD with reasonable pricess and that can handle a large number of writes.

[polarathene]

the biggest benefit that mechanical HDDs still have is longevity when compared to SSDs

Have you personally had any SSD fail on you? I've had a few HDDs do such, but yet to experience an SSD fail.

F2FS on the other hand is designed explicitly for flash storage, which is rather interesting, but I don't know how good it is. Also, I don't know if it is any useful since most flash storage devices have a Flash Translation Layer (FTL) which emulates a traditional storage device.

More geared at the flash storage that USB sticks usually provide or eMMC kind that you'd find on SBCs or phones. At least that was the original focus for F2FS if I recall correctly. Would be interesting to see how well it does/compares with the bigger SSDs.

[skeevy420]

Have you personally had any SSD fail on you? I've had a few HDDs do such, but yet to experience an SSD fail.

Nope. Haven't had one long enough for that to be a factor either. I have read horror stories of early adopters (both drives and file systems) and was thinking along those lines.

More geared at the flash storage that USB sticks usually provide or eMMC kind that you'd find on SBCs or phones. At least that was the original focus for F2FS if I recall correctly. Would be interesting to see how well it does/compares with the bigger SSDs.

That's what I've gathered about F2FS too so I'd be curious if Michael would test ext4 vs F2FS one of these days.

[zyxxel]

Have you personally had any SSD fail on you? I've had a few HDDs do such, but yet to experience an SSD fail.



More geared at the flash storage that USB sticks usually provide or eMMC kind that you'd find on SBCs or phones. At least that was the original focus for F2FS if I recall correctly. Would be interesting to see how well it does/compares with the bigger SSDs.

If I ignore the brand new disk that I managed to push off the table, I have had 4 mechanical disks fail in over 30 years. At this very moment, I have maybe 20 or 30 HDD powered. I have accumulated at least 500 years of power-on time with lots of HDD - but probably closer to 1000 years. But 500 years / 4 means I have had an average of 125 years of power-on time per HDD failure.

I have had 2 SSD fail. I might have accumulated 50 years combined runtime with SSD, which means I may have somewhere around 1 fail per 25 years.

1 of the 4 HDD failed so data couldn't be retrieved. Both of the SSD failed so data couldn't be retrieved.

Obviously, I still have too little combined runtime to have any good statistics of probability of failures. But SSD aren't the magically perfect devices that doesn't fail.

Depending on usage case, it really is quite easy to reach the rated write endurance of an SSD - especially since you need to take into account write amplification - where 1 GB total writes might represent 10 GB wear on the drive. A file system like BTRFS is at the bottom of the list when it comes to wear amplification - it produces huge amounts of disk writes.