Linux 4.16 File-System HDD & SSD Tests With EXT4/F2FS/Btrfs/XFS

[phoronix]

Phoronix: Linux 4.16 File-System HDD & SSD Tests With EXT4/F2FS/Btrfs/XFS

With the Linux 4.16 kernel release expected in just a matter of days, here are some fresh file-system benchmarks on this near-final kernel using a solid-state drive and hard drive while testing out the popular mainline file-system choices of Btrfs, EXT4, F2FS, and XFS.

http://www.phoronix.com/vr.php?view=26157

[elvenbone]

So, can it be concluded that ext4 remains the best compromise for consumer home PCs?

[Veto]

So, can it be concluded that ext4 remains the best compromise for consumer home PCs?

Depends... Given that most consumer home PCs run on SSDs, never serves web pages, only sporadically run databases and do few if any compilations, I would say no!
Instead a consumer home PC will benefit from fast reading, transparent compression and "Time Machine" like snapshotting.

I have changed to btrfs myself and see no reason to look back. However I still miss better desktop integration with the snapshot features.

BTW what is up with the extremely slow btrfs HDD sequential reads in Flexible IO?? I am running btrfs on several HDDs with no perceivable slowdown compared to previously running ext4.

[tillschaefer]

There too many oddities in this test, to derive any real world conclusions IMO.

For example: Why is the HDD faster than the SSD on the Compile-Bench initial create test? The results are much faster (up to 500 MiB/sec) than the sequential performance of the drive. Is the inital create completely fitting in the cache of the drives?

An then, there the sequential reads in the flexible IO tester for btrfs are unbelievable slow, but the random reads are missing for the same test on the HDD. And also random reads are given by IOPS and sequential reads as MiB/sec. And there is no other test that actually looks at the pure read performance of btrfs on the HDD....

I am not saying the numbers are wrong, I just do not get any pattern here, and it is therefore not really helping to draw any conclusions for the real world setting.

[schmidtbag]

So, can it be concluded that ext4 remains the best compromise for consumer home PCs?

To my understanding, EXT4 is supposed to be the most basic, reliable, and compatible Linux filesystem without being insecure. It doesn't specialize in anything, where it won't ever unanimously be the best option, but it's also never the worst choice. If you don't know what to pick, go with EXT4.

For me personally, I use BTRFS for a drive that I store my games on. I disabled things like copy-on-write while enabling a decent amount of compression. Data integrity/security and write speeds are largely irrelevant for this drive, while salvaging disk space and increasing read speeds is a high priority. Most games don't tend to load in large chunks of data during gameplay, so the added CPU usage and latency doesn't really hurt anything.

[RomuloP]

I'm trying to imagine what is so wrong with flexible IO or this seagate specific model when I get something like 80x the speed with a ST4000DM000-2AE166 with BTRFS.

[cjcox]

Other things. Ext4 can be shrunk, XFS cannot. You can rsync and get advanced ACLs (e.g. a samba area) with Ext4, can't get that with XFS. However, that testing I did from at least a year ago. But I think that's still true.

XFS is nice for growing, as it grows really fast, vs. Ext4. But you'll cry if you ever need to shrink. And if you use rsync to backup SMB, you might be frustrated if your backups are going to XFS.

[waxhead]

To my understanding, EXT4 is supposed to be the most basic, reliable, and compatible Linux filesystem without being insecure. It doesn't specialize in anything, where it won't ever unanimously be the best option, but it's also never the worst choice. If you don't know what to pick, go with EXT4.

For me personally, I use BTRFS for a drive that I store my games on. I disabled things like copy-on-write while enabling a decent amount of compression. Data integrity/security and write speeds are largely irrelevant for this drive, while salvaging disk space and increasing read speeds is a high priority. Most games don't tend to load in large chunks of data during gameplay, so the added CPU usage and latency doesn't really hurt anything.

Uhm.... your configuration is not possible so you might want to reconsider:
You can not disable copy on write and at the same time have compression : ( https://btrfs.wiki.kernel.org/index....e/btrfs%285%29 )

[marciosr]

Has someone tested straits?

[gadnet]

well any change to see a ZFS vs BTRFS showdown ?