Autodetect on SSD results in ashift=9 but this is obviously nonsense. Please check whether the drive is rotational or not before propagating this brokenness onto my pool.
ek4
All 4 comments
Do you propose to enforce some ashift for all non-rotatinf devices? I think it is very wrong assumption. If flash block us 32KB, it may have no big difference whether ashift is 9 or 12, both are equally bad, while bumping ashift to 12 may reduce space ZFS efficiency and increase disk and flash traffic without need.
amotin
on 6 Nov 2020
If there is more than one way to detect the underlying device sector geometries, perhaps ZFS should use all of the along with rotational vs ssd, and provide a warning/error if there are discrepancies, forcing the user to explicitly set a value based on research of the device?
gdevenyi
on 6 Nov 2020
Do you propose to enforce some ashift for all non-rotatinf devices? I think it is very wrong assumption. If flash block us 32KB, it may have no big difference whether ashift is 9 or 12, both are equally bad, while bumping ashift to 12 may reduce space ZFS efficiency and increase disk and flash traffic without need.
This is not totally correct: while ashift=12 does not means aligning to flash page nowadays, SSD's controllers (and FTLs) are way more optimized to manage 4K writes rather than 512B ones.
I would propose a much simpler approach: simply let zfs create ashift=12 pools by default, leaving the choice of 512B (or other values) at user requests. This will solve 99% of user issues, leaving other users/sysadmins the power to specify different ashift when it make sense.
Note: I propose the above for both rotational (ie: HDD) and non-rotational (SSD, Optane, etc) storage devices. This will even have the good side effect of removing the device blacklist once included (still present?) in zfs.
shodanshok
on 7 Nov 2020
As of 2017, NAND-flash SSDs are tuned for 4096-byte IOs. Matching the flash page size is unnecessary and ashift=12 is usually the correct choice.
Even if the code is still setting ashift based on legacy nonsense the docs have it straight.
ek4
on 8 Nov 2020
Related issues
marker5a
路
4Comments
Baughn
路
4Comments
jakeogh
路
3Comments
seonwoolee
路
3Comments
Greek64
路
3Comments
Most helpful comment
This is not totally correct: while ashift=12 does not means aligning to flash page nowadays, SSD's controllers (and FTLs) are way more optimized to manage 4K writes rather than 512B ones.
I would propose a much simpler approach: simply let zfs create ashift=12 pools by default, leaving the choice of 512B (or other values) at user requests. This will solve 99% of user issues, leaving other users/sysadmins the power to specify different ashift when it make sense.
Note: I propose the above for both rotational (ie: HDD) and non-rotational (SSD, Optane, etc) storage devices. This will even have the good side effect of removing the device blacklist once included (still present?) in zfs.