Re: btrfs write-bandwidth performance regression of 6.5-rc4/rc3

[Wang Yugui <wangyugui@e16-tech.com>]

Hi,

> On Tue, Aug 01, 2023 at 04:56:58PM +0800, Wang Yugui wrote:
> > > Odd.  What CPU are you using to test?  It seems like it doesn't set
> > > BTRFS_FS_CSUM_IMPL_FAST as that is the only way to even hit a potential
> > > difference.  Or are you using a non-standard checksum type?
> > 
> > The CPU is E5 2680 v2.
> 
> Is this in a VM and not passing through cpu flags?  What does dmesg
> say when mounting?  Norally it should say something like:
> 
> [   23.461448] BTRFS info (device vdb): using crc32c (crc32c-intel) checksum algorithm

This is NOT VM.

dmesg output:
[  250.596544] raid6: skipped pq benchmark and selected sse2x4
[  250.602836] raid6: using ssse3x2 recovery algorithm
[  250.612812] xor: automatically using best checksumming function   avx       
[  250.895573] Btrfs loaded, assert=on, zoned=yes, fsverity=no
[  250.905249] BTRFS: device fsid f5ebfdd6-6bf6-4c2b-b47b-79517bc00c8f devid 3 transid 6 /dev/nvme3n1 scanned by systemd-udevd (1726)
[  250.922155] BTRFS: device fsid f5ebfdd6-6bf6-4c2b-b47b-79517bc00c8f devid 4 transid 6 /dev/nvme0n1 scanned by systemd-udevd (1729)
[  250.935965] BTRFS: device fsid f5ebfdd6-6bf6-4c2b-b47b-79517bc00c8f devid 1 transid 6 /dev/nvme1n1 scanned by systemd-udevd (1724)
[  250.968268] BTRFS: device fsid f5ebfdd6-6bf6-4c2b-b47b-79517bc00c8f devid 2 transid 6 /dev/nvme2n1 scanned by systemd-udevd (1723)
[  251.070139] BTRFS info (device nvme1n1): using crc32c (crc32c-intel) checksum algorithm
[  251.079164] BTRFS info (device nvme1n1): using free space tree
[  251.089871] BTRFS info (device nvme1n1): enabling ssd optimizations
[  251.096921] BTRFS info (device nvme1n1): auto enabling async discard
[  251.104417] BTRFS info (device nvme1n1): checking UUID tree

Best Regards
Wang Yugui (wangyugui@e16-tech.com)
2023/08/01