Trying Out & Benchmarking Bcachefs On Linux 6.7

[phoronix]

Phoronix: Trying Out & Benchmarking Bcachefs On Linux 6.7

The biggest surprise this week so far with the Linux 6.7 merge window has been the landing of the Bcachefs file-system. Here is an early look at Bcachefs with Linux 6.7 and some preliminary benchmarks.

https://www.phoronix.com/review/bcachefs-linux-67

[Kjell]

Was really hoping the next round of FS benchmarks would include ZFS

[Michael]

Was really hoping the next round of FS benchmarks would include ZFS

OpenZFS doesn't build against Linux 6.7 git yet.

[skeevy420]

Moral of the story: When in doubt, pick XFS.

Here's to hoping Bcachefs becomes more performant.

Could this be just a case of less testing without compression enabled in regards to Bcachefs? I noticed that no compression was enabled. Perhaps LZ4 as the foreground compressor could affect things?

[fitzie]

that's some brutal fio results, but those were all direct io, which isn't typical of most applications. clearly bcachefs isn't optimized for those apps just yet. I know it seems pointless to add buffered fio tests, but I think it's more typical and I'd guess it would should a much more closer results. i would like to have seen those.

[Snaipersky]

Sounds like either you were formatting nvmeX rather than a namespace nvmeXnX, or you might have a bad block on your shiny u.3 drive. Does SMART have anything to say?

[Mathias]

BTRFS needs 1s to start xterm? 3 to start gnome-terminal? Both are <1s on BTRFS for me. (And I don't usually use both, so they weren't already cached and I'm on a 3yo medium end NVMe)

*edit* Michael clarified that there is a background I/O Mix running while performing the startup measurements. Of course I didn't test with that, so my results are obviously better. Thanks Michael

[chromer]

What we can see from benchmarks is that ext4 is currently far better than btrfs in performance matter.

[Danny3]

I was waiting for benchmarks for the new scheduler, but anyway, thanks for the benchmarks!

[Michael]

Sounds like either you were formatting nvmeX rather than a namespace nvmeXnX, or you might have a bad block on your shiny u.3 drive. Does SMART have anything to say?

I was trying it both ways when fiddling around with what the error might be. No smart errors and all other file-systems work fine.