Btrfs On 4 x Intel SSDs In RAID 0/1/5/6/10

[phoronix]

Phoronix: Btrfs On 4 x Intel SSDs In RAID 0/1/5/6/10

Earlier this month I published Btrfs RAID benchmarks on two HDDs but as some more interesting results are now Btrfs RAID file-system benchmarks when testing the next-generation Linux file-system across four Intel Series 530 solid-state drives. All RAID levels supported by the Btrfs file-system were benchmarked atop Ubuntu 14.10 with the Linux 3.18-rc1 kernel: RAID 0, 1, 5, 6, and 10 levels along with testing a Btrfs single SSD setup and a Btrfs file-system linearly spanning all four drives.

http://www.phoronix.com/vr.php?view=21104

[kgardas]

Please also test ZFS on Solaris 11.2/FreeBSD/Linux

Phoronix: Btrfs On 4 x Intel SSDs In RAID 0/1/5/6/10

Earlier this month I published Btrfs RAID benchmarks on two HDDs but as some more interesting results are now Btrfs RAID file-system benchmarks when testing the next-generation Linux file-system across four Intel Series 530 solid-state drives. All RAID levels supported by the Btrfs file-system were benchmarked atop Ubuntu 14.10 with the Linux 3.18-rc1 kernel: RAID 0, 1, 5, 6, and 10 levels along with testing a Btrfs single SSD setup and a Btrfs file-system linearly spanning all four drives.

http://www.phoronix.com/vr.php?view=21104

Hello,

with such nice setup it's really a pity not to test ZFS on Solaris 11.2, FreeBSD 10.x and Linux, don't you think? :-)

Thanks!
Karel

[oleid]

The comparison "Btrfs single + mdraid lvl X" vs "the same raid level in btrfs alone" would be interesting. Especially, since you can change the stripe size in mdraid -- in btrfs it's fixed to 64k AFAIK.

[liam]

Thanks for these tests, Michael.
There were some very interesting (nonlinear) results.
Is love to see two things added in future tests: mdadm raid for comparison, and CPU usage.
The scalability of btrfs should be improving, especially now that it's developers are working at a place where they have easy access to absolutely monstrous arrays.

[Michael]

Hello,

with such nice setup it's really a pity not to test ZFS on Solaris 11.2, FreeBSD 10.x and Linux, don't you think? :-)

Thanks!
Karel

Likely too much work without multiple premium users and/or donaters requesting such tests.

[r3pek]

I have been using a RAID 5 for quite some time now on 3 HDD drives (3 1TB seagate black).

I started with 1 black, 1 blue and 1 green (all 1TB), and then replace the blue and the green for 2 black and all went fine. The system had already suffered some forced power offs and the array was completely recovered without any loss (that I notived) and I think RAID on Btrfs (even 5/6) is pretty solid.

[gigaplex]

RAID 5/6 is still considered experimental for Btrfs but in the few days I've been running those configurations I haven't encountered any problems.

That's because you're testing performance under normal conditions, not reliability under abnormal conditions. The purpose of RAID 5/6 resiliency, and it's not fit for purpose there yet. A single disk failure can take down the whole array, making it not much better than RAID 0 in its current state.

[mufasa72]

Why are the RAID 1 reads so slow? You've got 4 sources you can read from that can be farmed out. RAID 0 and NON-RAID seems to be farming it out to the multiple sources.

[gigaplex]

Why are the RAID 1 reads so slow? You've got 4 sources you can read from that can be farmed out. RAID 0 and NON-RAID seems to be farming it out to the multiple sources.

Probably because the btrfs code doesn't yet farm it out in RAID 1 cases, even though it should. Might be worth filing a bug report in case someone else hasn't already.

[gigaplex]

Probably because the btrfs code doesn't yet farm it out in RAID 1 cases, even though it should. Might be worth filing a bug report in case someone else hasn't already.

Then again I'm not sure how non-RAID sequential reads are significantly faster than single disk, as if it's actually a linear span like Michael claimed, it won't have multiple sources to read from.