Linux ZFS?

[Gnome]

I noticed recently ZFS is now an option on Ubuntu?
Has anyone given this a spin and what are your thoughts?

This isn't a Linux vs FreeBSD, nor Ubuntu vs FreeNAS or anything like that thread.

Wondering out loud, I wonder if a Linux based device would offer more in the long run, as a NAS, than FreeBSD based device.
Not because Linux is a better or FreeBSD is worse.

Primarily it is about user base, or rather, the number of users. Linux has a lot more users.
Because of that much better driver support in my experience, which is one of the most annoying things about FreeBSD.
But some of the additional pains is the software. Linux has a lot of software and it gets updated fairly regularly and usually gets updated there first.

Thoughts?

Again this isn't a bashing thread, so I'm hoping this can be a civilized discussion :)

 

[danb35]

I noticed recently ZFS is now an option on Ubuntu?

ZFS has been an option on Ubuntu for some time, as well as on other flavors of Linux. It works well there too. If you want a Linux-based ZFS NAS, there are already other options on the market like OpenMediaVault and Rockstor (currently using btrfs, but will be switching to ZFS soon). Suggest you check them out.

Edit: Well, I thought OpenMediaVault used ZFS, but though their website lists a number of filesystems, ZFS isn't among them.

 

[Arwen]

I use Linux on the following;

• AMD64 quad core desktop
• AMD64 quad core small form facter, fanless, (for my media server)
• Intel64 dual core laptop

At one time I used Linux BTRFS on them for alternate boot environments and data validation, (but not mirroring, none had a second storage device).

About 3 years ago I changed the file system to ZFS on Linux, on my media server. It worked great. Later I changed my desktop. And eventually I changed my laptop. For the desktop and media server, I was able to mirror the OS to separate device. (Note I when I mean changed the file system, I mean all file systems including root on ZFS.)

Sometime in 2016 if I recall correctly, the OpenZFS project stablized the API so that userland programs can work with different versions of the kernel modules. Before this, if an update screwed up, it may have been necessary to boot to alternate media to fix the problem.

Lots of things work well for ZFS on Linux. Since ZFS 0.7.x, (I don't remember the exact sub-minor release), you can run some ZFS command options from a normal user account. (Solaris has always had this feature and current FreeBSD has it too.) I even played with at rest encryption a few months ago.

Over the years I wrote a script to create my alternate boot environments using ZFS. The script is not generic, as it expects a certain environment. But, for me, I can create a new boot environment in 30 seconds and then boot off of it. So if an update goes south, not a problem. Boot back to a known working snapshot & clone.

None of the Linux distros that support ZFS, make installing Linux on ZFS root easy. Kinda annoying. Wish Ubuntu did that. With a GUI or TUI. That would get more ZFS users.

Linux desktops and laptops should be protecting their user's data, (with mirroring if possible, but at least checksuming).

 

[danb35]

Linux has a lot of software and it gets updated fairly regularly and usually gets updated there first.

I'm still trying to figure out what the relevance of this (claimed) fact is to a NAS appliance. Samba/NFS/Netatalk/etc. are the same whether running under FreeBSD or Linux, no?

 

[Arwen]

@danb35,
In my opinion, Linux tends to have a faster, (and less tested & thought out), development cycle than most other free OSes.

Simple case in point, at rest encryption for ZFS is on Linux. But the company that is funding it appears to use a Solaris workalike.

That said, I do like that SSH is carefully updated on one OS, (one of the BSDs, FreeBSD?), then that version is ported. (I don't know exactly what they call it, or how it works, but it seems more well thought out. For a program that has security as the main component, I like that.)

 

[danb35]

In my opinion, Linux tends to have a faster, (and less tested & thought out), development cycle than most other free OSes.

...which would tend to suggest it isn't ideal for server applications. Though it probably isn't fair to paint Linux with such a broad brush; there are some distros (like RHEL/CentOS) that tend to be much more deliberate in their release cycles.

 

[Ericloewe]

I do like that SSH is carefully updated on one OS, (one of the BSDs, FreeBSD?)

OpenSSH is developed on OpenBSD.

 

[wblock]

The theory that more users means "more to offer" could be used to justify porting ZFS to Windows. A start on that was shown at the OpenZFS conference last week. It won't be mature for a while, possibly a long while.

FreeBSD has advantages for FreeNAS. For instance, FreeBSD has been running ZFS for something like a decade now. There's also the BSD license.

 

[Arwen]

...which would tend to suggest it isn't ideal for server applications. Though it probably isn't fair to paint Linux with such a broad brush; there are some distros (like RHEL/CentOS) that tend to be much more deliberate in their release cycles.

Yes, the server distros of Linux tend to be more conservative than the general wild Linux distros. The problem is that they have to back port new drivers, (like 10GBase-T, 25Gbps Ethernet, or 8Gbps / 16Gbps Fibre Channel), to old kernels that never had such devices. This costs money and time.

@Ericloewe, thanks for looking that up. I almost did a web search before I posted, but got distracted by something, (backups, or was it that horribly long Linux update? :).

On the subject of ZFS on Windows, holy cow batman! When I read that through the OpenZFS web site, I wondered how far it would go. I'd much rather see ZFS implemented as the root file system for Linux desktops & laptops, so we can get some protection. (Either mirror to 2nd device, or copies=2 for the user's home directory and critical files.) Not to mention allow alternate boot environments for easy software updates and roll backs.

 

[danb35]

I'm hoping this can be a civilized discussion

Were you planning to participate in the discussion?

 

[Ericloewe]

Were you planning to participate in the discussion?

Yeah, don't be shy, it's always interesting to see a somewhat more outsider take on things. Building insular communities is counterproductive, that's why I'm glad to see a lot of cross-pollination between the BSDs and Illumos distributions - it's easier because they're all more or less close cousins, but it's still positive.

That's something I like about ZFS, it runs pretty much on all extant major Unix-like OSes (OpenBSD is a notable, significant exception, but it's not due to outright incompatibility). I look forward to a stable ZFS on Windows (It'd be fantastic if Microsoft went "why are we wasting money developing our own OS if we can invest the same cash and get something that has been working for over a decade", but that's probably but a dream) so then I can say "all major extant desktop/server OSes". It helps break down barriers between OSes and between OS communities.

 

[Gnome]

I'm still trying to figure out what the relevance of this (claimed) fact is to a NAS appliance. Samba/NFS/Netatalk/etc. are the same whether running under FreeBSD or Linux, no?

Well agree or not, many people see a NAS appliance as both their personal storage and local server.
I draw the line at running my router (personally using PfSense) separate from my NAS.
But running other software on my NAS is definitely a must for me.

I feel like docker was a good move on Coral (on an otherwise problematic product) because of the ease it adds.
Now obviously you don't need Linux for this, but I think it would unburden the developers to focus on things other than building a framework to run dockers.

There was a post about by one of the current team members saying that the Coral code went so far as to build their own DHCP implementation in Python (or importing one).
Point being, as a developer, I feel like that effort could be better spent by focus on the core of a NAS while relying on others work for things like docker.

Now obviously I'm opening a can of worms here but I do get the impression the current FreeNAS developers are spending a lot of time as FreeBSD contributors also.

 

[Gnome]

The theory that more users means "more to offer" could be used to justify porting ZFS to Windows. A start on that was shown at the OpenZFS conference last week. It won't be mature for a while, possibly a long while.

FreeBSD has advantages for FreeNAS. For instance, FreeBSD has been running ZFS for something like a decade now. There's also the BSD license.

Hmm, I haven't used Windows for years now but certainly I can see some users being very excited by that idea.

I'll say however that when it comes to automation, scripts, etc. I've found much better resources online for Linux. But that was when I made the switch to using *Nix based servers around 10 years ago.

 

[Gnome]

Were you planning to participate in the discussion?

Sorry, I've been busy at both work and home :p

That's something I like about ZFS, it runs pretty much on all extant major Unix-like OSes (OpenBSD is a notable, significant exception, but it's not due to outright incompatibility). I look forward to a stable ZFS on Windows (It'd be fantastic if Microsoft went "why are we wasting money developing our own OS if we can invest the same cash and get something that has been working for over a decade", but that's probably but a dream) so then I can say "all major extant desktop/server OSes". It helps break down barriers between OSes and between OS communities.

Agreed, if Oracle realized what this would buy them, they wouldn't resist it so much.

There is a lot to be gained by having a filesystem that runs on every single operating system.
And being a company that delivered that is a very powerful idea IMO.

We don't have anything pretty much universal other than FAT16/FAT32 (I may be a bit out of touch here, but I believe this is still the case?)
The world has to move on

 

[danb35]

I'll say however that when it comes to automation, scripts, etc. I've found much better resources online for Linux.

That's probably a fair assessment. But most of those will drop right in to FreeBSD, as the languages used and the tools are pretty much the same (although there does seem to be gratuitous incompatibility in the GNU versions of some common tools).

 

[Ericloewe]

There was a post about by one of the current team members saying that the Coral code went so far as to build their own DHCP implementation in Python (or importing one).

Well, that's not a problem with FreeBSD (or specific to it) as much as a limitation of existing Unix-style software. Even then, the approach taken was dubious.

Agreed, if Oracle realized what this would buy them, they wouldn't resist it so much.

Oracle is irrelevant here. Their ZFS is probably inferior by now and it's definitely incompatible. Their only relevance is that they might try to sue Microsoft if they were to ship ZFS, but they'd have a very hard time.

 

[danb35]

I do get the impression the current FreeNAS developers are spending a lot of time as FreeBSD contributors also.

What does this have to do with anything in this thread?

 

[danb35]

Oracle is irrelevant here. Their ZFS is probably inferior by now

Encryption, though...

 

[Ericloewe]

Encryption, though...

Well, OpenZFS has encryption now, too.

 

[danb35]

Well, OpenZFS has encryption now, too.

In production-ready code? I'd understood it was under development, but not ready for production yet.