The Butter goes on top (Safe BTRFS raid5/6)

I think a number of these more recent patches have been submitted for kernel 4.9, while 4.8 and new btrfs-progs have just been released:

1 Like

I have been following the Btrfs mail list some more.

And even though work is being done, it seems that production ready RAID56 still needs a lot of work.

Especially this thread has a lot of information about the challenges, and the complexity of fixing this problem. Its a long thread with a lot of technical talk.

https://mail-archive.com/linux-btrfs@vger.kernel.org/msg58231.html

Basically the current way Btrfs and CoW works is not well suited for RAID56, but some very knowledgeable people are discussing solutions, and it seems possible to fix the problems.
But I think a lot of work and testing needs to be done.
The current implementation is very buggy.
Described this way by one of the people discussing the problems:
“The current btrfs raid5 implementation is a thin layer of bugs on top of code that is still missing critical pieces.”

1 Like

This means for raid5/6, go with ZFS for the time beeing if i am to read this correctly

Perhaps.

I dont think ZFS in Rockstor is going to happen.

Even though ZFS is more mature in some ways, it lacks the flexibility and the low system memory requirements that BTRFS has.
And Rockstor is build being a BTRFS centric system.

But for people wanting to run RAID5/6, a ZFS based NAS solution such as Freenas would be the better option.

Flexibility I agree with you @KarstenV, for memory, there is this missconception that one will need 1GB Ram per TB data, which is true if you use De-Duplication.

For a regular homeNAS, nah, go with as much as possible (same as for Rockstor) and be happy.
According to this http://louwrentius.com/things-you-should-consider-when-building-a-zfs-nas.html
Freenas requires 4GB (ideally 8) to work, as Freenas lives in memory.

From the link (Raidz with 5 drives):

For maximum performance, you should have enough memory to hold 5 seconds worth of maximum write throughput ( 5 x 400MB/s = 2GB ) and leave sufficient headroom for other ZFS RAM requirements. In the example, 4 GB RAM could be sufficient.

The licensing issue has been discussed and also Rockstor beeing built as a BTRFS centric system is a valid point. I love Rockstor, but as the BTRFS Raid 5/6 is so unstable, I have had to look other places, but I dont feel like home as I do with Rockstor.

I have run a Freenas setup in 4Gb with little or no trouble except an occasional slowdown (as I remember it, it used its swap partition a lot?), so I agree to some degree that it is not that memory dependant.

But my current rockstor system, with 7,8TB storage and a plex media server running, shows memory usage at about 1Gb, with the rest being free / buffer/cache. No swap-space used.
So my Rockstor setup could probably live happily in 2Gb ram, with little to no performance hit (perhaps a small amount of swapping).

I would not have been happy running my Freenas on that kind of setup.

But ram is relatively cheap, so this is no major hurdle.

I agree that Rockstor has a lot of positives going for it. So I really hope that these RAID56 issues get resolved as quick as possible as it would really be good for Rockstor as a Freenas alternative.

1 Like

So do I, it’s very unfortunate that we have to make do with an incomplete filesystem. On the other hand, there is time to polish our support of btrfs features to perfection before the mobs come running to Rockstor for being a mature btrfs system. :wink::wink: