Btrfs Gets Improved Performance On Linux 3.8 Kernel

[phoronix]

Phoronix: Btrfs Gets Improved Performance On Linux 3.8 Kernel

The Btrfs file-system updates for the Linux 3.8 kernel will feature performance improvements among other work. Btrfs RAID 5/6 support is also to be published this week...

http://www.phoronix.com/vr.php?view=MTI1NTY

[benmoran]

....raid5/6 is being rebased against the device replacement code. I'll have it posted this Friday along with a nice series of benchmarks.

That's awesome. I'm really looking forward to testing out RAID6.

In terms of line count, most of the code comes from Stefan, who added the ability to replace a single drive in place. This is different from how btrfs normally replaces drives, and is much much much faster.

Does anyone have any more information about this? Currently, when you tell btrfs to remove a drive, it rebalance the data first (by moving the data to other drives in the array). This takes a while with big/full disks, since the data has to be written. How is the new method different? Instead off moving data off of the old disk, and then back on to the new disk, does it simply produce new stripes on the new drive, and not bother copying anything off first? I'm curious about this, but couldn't find much info. Anyone know?

[Veerappan]

That's awesome. I'm really looking forward to testing out RAID6.



Does anyone have any more information about this? Currently, when you tell btrfs to remove a drive, it rebalance the data first (by moving the data to other drives in the array). This takes a while with big/full disks, since the data has to be written. How is the new method different? Instead off moving data off of the old disk, and then back on to the new disk, does it simply produce new stripes on the new drive, and not bother copying anything off first? I'm curious about this, but couldn't find much info. Anyone know?

Disclaimer: I don't know how it actually does this.

If it's similar to how ZFS handles a drive replacement, it could be that it just copies the data byte-for-byte from the drive being replaced to the one that's replacing it (if the old drive is still available). Depending on if you've got a redundant disk set, this could still be possible with a failed disk in the mix (just use the redundant copies as the source). I don't see why you'd have to modify the data on the drives that are NOT being replaced.

[LasseKongo]

That's awesome. I'm really looking forward to testing out RAID6.

Are we really sure it is in ? I can´t see anything on the BTRFS devel mailinglist, and the latest GIT snapshot of the tool I downloaded 2 days ago
still misses support for RAID5/6.
I´m also eager to start testing this.