I noticed that, starting with some update, the FSYNCS in pveperf has become very slow.
In the past, I was getting around 9,668.38 with a single SAS disk connected to the 9500-16i, but just a moment ago it dropped to 4,532.19—I noticed the speed had been cut in half.
*Only SAS devices are affected. KCD6XLUL960G (U.2) devices connected to a PCIe port are not affected. It is unclear whether KCD6XLUL960G (U.2) devices connected via SATA or through a tri-mode 9500-16i are affected.
As for the mirrored disk, it wasn’t even reaching 2,000.
To be honest, I don’t even know which update caused this. I hadn’t run pveperf since the end-of-year holidays, so I didn’t notice it until now.
Since the only changes I’d made were the regular PVE updates and the 9500-16i firmware update, I first rolled back the firmware to versions P38, P37, and P34, but this didn’t improve the situation.
I also pinned the kernel to 6.14, but this didn’t help either.
I tried various things, and the only settings that brought the values back close to their original levels were the following—though I don’t recall changing them. I don’t have such settings in `modprobe.d`.
Could these settings have been automatically changed by a ZFS update or something similar?
・Previously Saved Results
・Results When a Decline Is Confirmed
・After changing the settings (zfs_nocacheflush)
・The KCD6XLUL960G (U.2) device consistently produces the same results.
In the past, I was getting around 9,668.38 with a single SAS disk connected to the 9500-16i, but just a moment ago it dropped to 4,532.19—I noticed the speed had been cut in half.
*Only SAS devices are affected. KCD6XLUL960G (U.2) devices connected to a PCIe port are not affected. It is unclear whether KCD6XLUL960G (U.2) devices connected via SATA or through a tri-mode 9500-16i are affected.
As for the mirrored disk, it wasn’t even reaching 2,000.
To be honest, I don’t even know which update caused this. I hadn’t run pveperf since the end-of-year holidays, so I didn’t notice it until now.
Since the only changes I’d made were the regular PVE updates and the 9500-16i firmware update, I first rolled back the firmware to versions P38, P37, and P34, but this didn’t improve the situation.
I also pinned the kernel to 6.14, but this didn’t help either.
I tried various things, and the only settings that brought the values back close to their original levels were the following—though I don’t recall changing them. I don’t have such settings in `modprobe.d`.
Code:
zfs_nocacheflush=1Could these settings have been automatically changed by a ZFS update or something similar?
・Previously Saved Results
Code:
・sas KIOXIA KPM5XMG400G
HD SIZE: 358.47 GB (rpool/ROOT/pve-1)
FSYNCS/SECOND: 9668.38
No results found for husmm3280ass201・Results When a Decline Is Confirmed
Code:
・sas KIOXIA KPM5XMG400G
HD SIZE: 357.67 GB (rpool/ROOT/pve-1)
FSYNCS/SECOND: 4532.19
・sas husmm3280ass201 (zfs mirror)
HD SIZE: 630.66 GB (zfs_husmm3280ass201)
FSYNCS/SECOND: 1387.50・After changing the settings (zfs_nocacheflush)
Code:
・sas KIOXIA KPM5XMG400G (zfs Single disk)
HD SIZE: 357.67 GB (rpool/ROOT/pve-1)
FSYNCS/SECOND: 9035.55
・sas husmm3280ass201 (zfs mirror)
HD SIZE: 630.65 GB (zfs_husmm3280ass201)
FSYNCS/SECOND: 3577.64・The KCD6XLUL960G (U.2) device consistently produces the same results.
Code:
・U.2 KIOXIA KCD6XLUL960G
HD SIZE: 860.53 GB (rpool/ROOT/pve-1)
FSYNCS/SECOND: 16415.25
Last edited: