EXT4 Seeing Work To Speed Up Mount Times For Large File-Systems

[phoronix]

Phoronix: EXT4 Seeing Work To Speed Up Mount Times For Large File-Systems

Patches are pending that can sharply speed-up mount times of large EXT4 file-systems...

http://www.phoronix.com/scan.php?page=news_item&px=EXT4-2020-Faster-Mount-Times

[StarterX4]

"mount process w/o the patch couldn't complete in half an hour (according to vmstat it would take ~10-11 hours). with the patch applied mount took ~20 seconds."
Weird bug.

[George99]

Weird bug.

Or a clever trick. Like adding integer numbers from 1 to 100 by hand or just group them in pairs and get 50*101.

[Azrael5]

How long for this improvement to be available?

[starshipeleven]

How long for this improvement to be available?

ask your own distro maintainers if they plan to backport this

Otherwise it will be in the next kernel version, as usual.

[kylew77]

Is there much of a use case for 100+ TB arrays formatted in EXT4? Wouldn't most work loads of that size be ZFS or BTRFS?

[starshipeleven]

Is there much of a use case for 100+ TB arrays formatted in EXT4? Wouldn't most work loads of that size be ZFS or BTRFS?

That's just 10x 10TB drives, and there is plenty of workloads on servers where you don't care about ZFS/btrfs, like for example for databases (that are CoW and checksummed already internally).

[smitty3268]

I wonder how slow dm-slow is.

[kylew77]

That's just 10x 10TB drives, and there is plenty of workloads on servers where you don't care about ZFS/btrfs, like for example for databases (that are CoW and checksummed already internally).

Thank you. One thing my IT education has been light on is data bases so I didn't know that databases are already checksummed.

[yoshi314]

"mount process w/o the patch couldn't complete in half an hour (according to vmstat it would take ~10-11 hours). with the patch applied mount took ~20 seconds."
Weird bug.

i have a not that big XFS filesystem at work, with a very heavy file creation/deletion traffic. and unmounting it takes ~15 minutes - most likely to make sure journal is up to date.

so cases like these are not that uncommon, regardless of fs.