Skip to contentSkip to conversation

comment about 'btrfs: add ssd_metadata mode'

DISCUSSION Goffredo Baroncelli 1 participant 1 message Apr 2, 2020 2 messages not in the archive Btrfs file system

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw
Quoted 4 lines · Wang Yugui

Hi, Goffredo Baroncelli

'btrfs balance' after 'mkfs.btrfs' seem OK to make sure not any HDD will be used as metadata.

Yes this is the expected behavior.

Quoted 16 lines

test env:
kernel:5.4.29 + patch(btrfs: add ssd_metadata mode)
/dev/sdb hdd
/dev/sdc ssd

test script:

	mkfs.btrfs -m single -d single /dev/sdc1 /dev/sdb1 -f
	mkdir -p /test &&
	mount /dev/sdb1 /test  || exit $?

	btrfs fi usage /test
	btrfs balance start --full-balance /test
	btrfs fi usage /test

output of the script

[...]

Quoted 18 lines

+ btrfs balance start --full-balance /test
Done, had to relocate 3 out of 3 chunks
+ btrfs fi usage /test
Overall:
Device size: 605.50GiB
Device allocated: 2.03GiB
Device unallocated: 603.46GiB
Device missing: 0.00B
Used: 640.00KiB
Free (estimated): 604.46GiB (min: 604.46GiB)
Data ratio: 1.00
Metadata ratio: 1.00
Global reserve: 3.25MiB (used: 0.00B)

Data,single: Size:1.00GiB, Used:512.00KiB

    /dev/sdb1       1.00GiB

Metadata,single: Size:1.00GiB, Used:112.00KiB

Quoted 8 lines
    /dev/sdc1       1.00GiB

System,single: Size:32.00MiB, Used:16.00KiB

    /dev/sdc1      32.00MiB

Unallocated:

    /dev/sdb1     418.19GiB
    /dev/sdc1     185.28GiB

If the /dev/sdc1 is an SSD, yes my code works as expected :-)

Quoted 12 lines

Best Regards
王玉贵
2020/04/02

Quoted 7 lines

Hi, Goffredo Baroncelli

Thanks for the nice ideal of 'btrfs: add ssd_metadata mode'.

a comment:
Should we add ssd_metadata mode to mkfs.btrfs too? If so, we can let the first chunk works as expected too?

Having the first chunk correctly allocated, doesn't really matter. As you can saw a simple "btrfs bal" rearrange the chunks. Updating mkfs.btrfs is not so urgency [*].

This patch is an experiment to have some feedback. There are a lot of open points:
- it is really useful ? I expect that in few years all disks will be SSD
- there are a lot of open point about what would returns (e.g.) "btrfs fi us": still is it correct to return as free space the sum of the free space of each disks ?
- increasing the speed of reading/writing metadata, how impact to the overall speed of the filesystem ? I expect no too much. If I find some spare time, I would measure the upgrading time of a debian from stable to unstable in the following scenarios:
- ssd
- ssd+hdd
- hdd

Quoted 4 lines
Quoted 4 lines

Best Regards
王玉贵
2020/04/02

BR
G.Baroncelli

Quoted 11 lines
Quoted 5 lines

--------------------------------------
北京京垓科技有限公司
王玉贵 [email protected]
电话:+86-136-71123776

--------------------------------------
北京京垓科技有限公司
王玉贵 [email protected]
电话:+86-136-71123776

[*] Anyway, I am inclined to think that the mkfs.btrfs should put the drive in a "initial state" only. The building of the files-system should be done only in the kernel the first time that it see a disk in a "initial state". This would avoid to have a lot of user space code which deals with the internal structure of a filesystem. There are a lot of repetition between the btrfs user space code and the btrfs kernel space code.

Signature · 2 lines
gpg @keyserver.linux.it: Goffredo Baroncelli <kreijackATinwind.it>
Key fingerprint BBF5 1610 0B64 DAC6 5F7D  17B2 0EDA 9B37 8B82 E0B5