I see that apparently one can now set volumes on the NAS to use Btrfs (currently via the Data Protection Manager app but apparently native in the upcoming DSM6.0)

So, is this a good thing and should I be using it over ext4?

I can see the advantages of snapshots to replace the Time Backup app (which I have found problematic), but is Btrfs ready for prime time? I was under the impression there were still some teething troubles (based on some recent posts here).

Any opinions?

Btrfs is being used "for prime time" all over the place.

My own reservation about it, is that we know very little about resilience in the face of power failures and other abrupt cutoff events.

The tools and experience with ext2/3/4 over the decades have resulted in a very robust solution that has yet to ever lose any data for me. Dunno about Btrfs.

Cheers

We've seen severe performance issues when hitting btrfs hard (creating/destroying hundreds of containers with layered filesystems with quotas) that made us move to aufs. Not relevant in the NAS use case, but I'm a little hesitant to buy into btrfs.

For the record, I'm still sad that Apple killed ZFS and still hasn't brought up a proper alternative. TimeMachine is a kludge, in comparison, but it's currently doing the job for me.

Thankfully Apple's ability to murder tech still remains something they can only do to their own stuff. ZFS still exists and can be used, however Oracle still retains enough control to make it unsuitable for Apple to bundle it in OS X as previously planned.

Of course, Oracle also controls Btrfs, right?

Meanwhile, it seems that the NetApp patents in question are at least starting to expire. Their earliest asserted patent appears to have expired last year. (But the rules on expiration are non-trivial, so if it's not dead now, it will die very soon.) The later NetApp patents are still in force.

If Apple wanted to revisit the issue, they could almost certainly afford to mount a vigorous defense against patent infringement, especially if they deliberately engineered a filesystem with the specific purpose of working around the various patent requirements. What's sad, however, is that I don't think Apple really gives a shit about instant snapshots and other such things. Those are things you really want on file servers, which Apple once sold but doesn't sell any more.

Peter, I've not had hands on experience with Btrfs at this point. Though I'd trust it on a NAS a bit more since NetGear already switched the ReadyNAS line over back in 2013. That at least added a larger group of "beta" users and likely led to a lot of improvements and some stability time. While the engineering group mostly changed, they still continued their general interaction with the rest of the community to help develop and improve the situation. It's similar to their past direct work on Netatalk which helped keep that project alive, while other NAS vendors of the era were simply taking code without contributing back.

At some point late this year I'll likely hack my older ReadyNAS unit to run the newer OS with Btrfs support to try. Doing so requires me to default the system and put data back on it, and I'm not quite prepared to do that currently. Though my personal goal is still to get to ZFS, and long term I'll be trying to hack the NAS to switch to FreeBSD to accomplish this. Or get a newer unit that is FreeNAS ready out of the box.

The core advantage of ZFS and Btrfs is still better data integrity then any other filesystem can offer, since it can check the data and not just filesystem metadata at any time. The rise of IO speeds and compute power has helped enable this without the drastic performance hits such required back on some proprietary SANs I worked with a decade or so ago.

I assume this was in reply to me Dan and not Peter directly.


Yep, and up front it was licensed under GPL (v2 I believe thankfully, since it's part of the Linux kernel).

This will still keep Apple away from Btrfs. Likely due to them already being burned by GPLv2 projects deciding to flip to GPLv3, and also the general toxicity that the GPL/GNU crowd shows towards Apple. All while forgetting they could have had LLVM. Good writeup here in why I hope this unnecessary toxicity cools off soon, from someone on the GNU side https://lists.gnu.org/archive/html/emacs-devel/2015-02/msg00516.html

For Oracle, their move makes sense to me. Their work on Btrfs on the Linux/GPL side, and keeping ZFS going under the CDDL allows them to advise their customers of the Oracle DB to use these filesystems. Thus helping the data integrity code in the DB side benefit from a data integrity filesystem.


All good to see at this point. Oracle and NetApp did seem to reach a good settlement ahead of this exploding in the courts. Sun didn't have the runway to really defend themselves. As much press was dedicated to the Oracle/Google patent fight and the anti-Oracle feeling it stirred up, not as much attention seemed to be dedicated to this other situation. Oracle has done a lot of harm in their own ways, however their good is often forgotten. I prefer their guidance of Hudson over the mess that Jenkins has turned into. Right off the bat, Hudson post fork went to a sensible version number scheme that helped with plugin compatibility, Jenkins today remains a mess in that area. Anyhow, slight tangent.


Patent wise around 2006-2007, Apple probably wanted to save their own lawyers bandwidth for establishing the iPhone/iPad patents, instead of defending a filesystem written by another company. And since the launch of the iPhone, they've been quite busy with incoming and outgoing lawsuits. Factoring in Job's intense focus and his known terminal condition by then, I'm not surprised in hindsight that ZFS in OS X was cut.

I'm pretty sure Apple gives a lot of good shits, their cafeteria has a lot of healthy offerings including high fibre foods. From a purely outside view of watching Apple carefully, they likely has some plan to improve the filesystem on OS X/iOS/tvOS/WatchOS within a few more years. I'm not sure if it would go to ZFS/ZFS like directly. A variant worked into the growing CoreStorage efforts that lacks some ZFS benefits while also cutting way down on overhead would be my guess currently. They do still see a number of support cases and issues due to flaws in HFS+ in all their consumer products.

Where their ZFS like feature work is also going on is in their cloud offerings. iCloud Photo Library, iTunes Match, and iCloud Drive all enable people to keep the master copy of data server side. This allows it to possibly be at rest on a bitrot detecting filesystem or other bitrot detection setup. Any client of the data be it OS X or iOS gains recovery from bitrot or other errors by allowing a fresh copy to be downloaded. Snapshots also happen here, allowing 30 days of recovery currently. Users still benefit from these features without their local filesystem supporting it.

It's a pretty tight secret on how Apple runs their own datacenters, though I'd guess they are running a BSD kernel/OS and quite possibly running ZFS. Swift mixes into their longer term plans in the datacenter I'd bet too. IMHO, their own datacenters and internal working knowledge on cloud architectures is still a little behind their strongest competitors. Though they seem to be learning quick through some key acquisitions, and hosting parts of iCloud on other cloud platforms like AWS and Azure. Apple's not enough of a "cloud only" company to make me think that these efforts are hindering improvements client side in the early Cook era.

Yeah, sorry. I get the linear view of the BBS and just started typing in the box at the bottom.

Anyway, one other useful data point is the way Apple engineered its battery pack for iPhones, perhaps to avoid Mophie's patents. Of course, you'll never get anybody from Apple to confirm that this was the plan, but it all seems to make sense.

Consequently, if and when Apple decides they want a proper client-side copy-on-write snapshotting filesystem solution, they'll make it happen.

Another reason to look elsewhere for useful stuff that Apple may not deem cool enough for the devoted Apple iConsumer. At work running 3 mirrored ZFS storage systems on OmniOS (http://omnios.omniti.com/) hosting about 64TB and, allowing for OpenSolaris learning curve, a super outcome and value. Farewell EMC, NetApp, Apple

We were gonna redeploy ZFS services to FreeBSD, but their ZFS implementation still suffers from a 16-group-limit-for-NFS problem.

Jim

Absolutely. TimeMachine would greatly benefit from instant snapshots, as that it basically what it currently has to implement at the moment on top of HFS+ in a horribly hacky way.

Also all the iOS devices would greatly benefit from it, what better way to deal with upgrades in a rollbackable-if-it-goes-wrong way.

My thoughts too Tom, to some extent. I would be more inclined to use btrfs given the weight and reputation of some big name business NAS vendors such as Synology and NetGear behind it.

Currently, my thoughts are to wait until DSM6 comes out of beta, then upgrade my DS710+ to a DS716+.

As mentioned previously, my main motivation is the snapshots which I am hoping will be a more robust and granular method of 'versioning' than Time Backup. I believe this part of btrfs is quite well tested and stable now.

Good point Mark, fortunately in my location the power is quite reliable although we get a fair few electrical storms in the summer which could be a problem but we have been lucky so far (touch wood!). The critical data is also backed up off-site nightly.

Very cool. Personally, my former DEC/Compaq/HP storage days has me applauding the farewell

Apple dipping their toe into that space again during the second Job's era was interesting to watch. Their 1U server designs were a good first attempt, with some major flaws. They seemed to address those quicker then some other companies. Back then, I was helping a Mac reseller understand them, along with the weaknesses that could be shored up with some 3rd party equipment.

I was curious to see where they could go with SAN technology. My particular Mac Pro that is still running today had a lot of options for adding fibre cards and the XSan software when it was bought in 2008. The XServer RAID though only really got to the basic level that the MSA1000 did for Compaq.