btrfs RAID0 3 disks

[harmonyp]

I came across the following on creating btrfs. I am curious what do you guys suggest if I want performance? What are the pros/cons for each

# Create a filesystem across three drives (metadata mirrored, linear data allocation)
mkfs.btrfs -d single /dev/sdb /dev/sdc /dev/sdd

# Stripe the data without mirroring, metadata are mirrored
mkfs.btrfs -d raid0 /dev/sdb /dev/sdc /dev/sdd

 

[Stefan_R]

"single" means the data is allocated in a linear fashion. That is, if all your three drives are 1TB in size, the first terabyte of data would all go to sdb, then the next to sdc, then sdd. This means that if a drive fails, data recovery might be easier, as most files are stored exclusively on one drive. IMO this is a minor benefit though, as you'll still lose a lot. The only reason I can see this being useful if your drives are of unequal size, since you can use all of it in this mode.

"raid0" is a regular RAID0, meaning your data is striped. AFAIK this happens in 1GB increments on btrfs, but for files of different applications etc... this will provide a speed benefit, as multiple drives are accessed at the same time. It may also increase longevity, as it's not just one drive getting hammered all the time. With varying size disks however, you will lose some space in this mode.

 

[harmonyp]

"single" means the data is allocated in a linear fashion. That is, if all your three drives are 1TB in size, the first terabyte of data would all go to sdb, then the next to sdc, then sdd. This means that if a drive fails, data recovery might be easier, as most files are stored exclusively on one drive. IMO this is a minor benefit though, as you'll still lose a lot. The only reason I can see this being useful if your drives are of unequal size, since you can use all of it in this mode.

"raid0" is a regular RAID0, meaning your data is striped. AFAIK this happens in 1GB increments on btrfs, but for files of different applications etc... this will provide a speed benefit, as multiple drives are accessed at the same time. It may also increase longevity, as it's not just one drive getting hammered all the time. With varying size disks however, you will lose some space in this mode.

Thanks for the input, RAID0 it is then although I have read single are in 64K chunks.

What do you think about putting the metadata as RAID1?

mkfs.btrfs -m raid1 -d raid0 -L My-Storage /dev/nvme0n1 /dev/nvme1n1  /dev/nvme2n1

good or bad for performance? I am expecting 5x more writes than reads generally.

 

[Stefan_R]

For a RAID0, metadata mirroring doesn't make too much of a difference, both in terms of consistency and performance. It might help recover data from a slightly damaged disk.