We’re a little lacking in performance, a 342.95MB/s Raid10 isn’t great for large libraries. Lumaforge’s Jellyfish can allegedly do 600MB/s+ with a Raid6+0 of rust drives.
https://www.45drives.com/wiki/index.php?title=RAID_01
Would be nice to have some options for those who need more speed than resilience. Perhaps rockstor support for mdadm configs?
Update: Based on some testing… raid10 is raid01 in btrfs. My single disks get 119.20 MB/sec, while a 6 disk RAID10 gets 342.95 MB/sec.
Rockstor is a BTRFS based system, not designed nor purposed for the use of mdraid.
While mdraid can be put under or even on top of BTRFS, that is against the ideology of Rockstor.
BTRFS is not built for speed. If I/O performance is what you’re guaging your interest on, you will likely be disappointed.
However, it is very good at allowing users to achieve some level of redundancy regardless of what disks they have available. (IE: If one has 3 1Tb drives, a 500Gb and a 750Gb, you can still achieve a redundant RAID with 2624Mb of usable space.)
Is also supports CoW and snapshot capability.
Think something more like ZFS, but lacking ZFS’ confusion about pool expansion, which requires creating VDEVs consisting of one or more disks, and adding them to the pool - Whereas BTRFS can arbitrarily add or remove disks from a pool (assuming sufficient space after removal).
BTRFS also less confusion about redundancy. AFAIK ZFS requires multiple devices per VDEV in order to achieve redundancy for that VDEV, regardless of the pool layout - whereas with BTRFS in an appropriate RAID format, any disk can be lost without incurring data loss)
Actually, there’s something interesting about BTRFS RAID1: It’s not a traditional RAID1 array in the sense that the same data is replicated to every drive. Instead, BTRFS RAID1 just means “keep two copies of every piece of data”. So, if you have four drives in a RAID1, every chunk of data you copy to it will exist in full on two disks. That doesn’t sound interesting, until you realize most files are several chunks long, so if you have a big file it could distribute it to all the disks in the whole BTRFS, with every chunk of data on at least two disks. That means read operations can be much faster than a single disk.
Definitely true on the RAID description. I’ve heard BTRFS RAID described in many ways (commonly RAID 1E), but I think your description is about the best.
I will mention however that I believe the processing overhead and early stages of development of BTRFS aren’t conducive to the read speed increases you mention. I have 8 disks (4x3Tb and 4x4Tb) and generally see lower access times than I typically do on similar MDRAID10 systems.