Oracle ZFS man calls for Big Red to let filesystem upstream into Linux

Oracle storage architect has called for Oracle to make the ZFS filesystem a first class part of Linux and says conversations have taken place within Big Red to consider the possibility. Speaking at the OpenZFS Developer Summit, Maybee said the decline of on-premises storage means ZFS' future inside Oracle is uncertain. Oracle' …

[Anonymous Coward]

Not going to happen

People talk about a lot of things inside Oracle, but forget about it happening - legal will never allow it. The lawsuit with NetApp is one huge reason. The other reason is Red Hat. Red Hat hate Oracle so much now they will never allow it to go upstream to them and would likely exclude it if Oracle got it in upstream from Red Hat. Unfortunately, Oracle poisons or pisses in every well it comes across.

Wishful thinking by developers after a new home for their technology doesn't count for much inside Oracle.

[fobobob]

Re: Not going to happen

If you could choose one to excise from the Linux ecosystem, which would you pick: Oracle, or Pottering?

[Anonymous Coward]

Re: Not going to happen

"If you could choose one to excise from the Linux ecosystem, which would you pick: Oracle, or Pottering?"

Can we choose both?

[PNGuinn]

Re: Not going to happen @fobobob

How about chucking him down the next well that Oracle's about to p*ss down?

[CrazyOldCatMan]

Re: Not going to happen

excise from the Linux ecosystem, which would you pick: Oracle, or Pottering?

Yes.

[Mage]

Re: Not going to happen

Retire Pottering

Oracle at least are only motivated by profit.

[Anonymous Coward]

Re: Not going to happen

If you could choose one to excise from the Linux ecosystem, which would you pick: Oracle, or Pottering?

Lets see, a choice between a company ruled by dickheads that *still* manage to, on occasion, produce something useful, or the person who invented PulseAudio and systemD...

[TVU]

Re: Not going to happen

"If you could choose one to excise from the Linux ecosystem, which would you pick: Oracle, or Pottering?"

Answer = a) preferably both and b) as soon as possible.

[thames]

Re: Not going to happen

The suggestion seems to be to get ZFS in Linux so that Oracle could sell that and phase out Solaris. If Red Hat wants nothing to do with ZFS, that would be so much the better so far as Oracle was concerned as they then wouldn't have that competition from them.

The point is to tap into the innovation and development that is happening on Linux, together with its far bigger base of existing software. IBM has gone that route with their newer line of mainframes.

Given the trends with respect to Solaris development, I won't be surprised to see Oracle go this way eventually.

[Anonymous Coward]

Re: Not going to happen

They don't need to do anything else to phase out Solaris, they've already effectively killed it.

Oracle needs Red Hat because they're downstream from Red Hat as they want all the Red Hat benefits and compatibility (at minimal cost), so Oracle won't go it alone.

Oracle is so focused on their cloud services, external Linux distributions are definitely a secondary or tertiary consideration for them.

[Mage]

Re: Solaris development

I heard Solaris development is a little slow these days? Rumours of it even ending.

It was a must have when you couldn't get a decent spec PC and Linux was in infancy and you wanted a Telco / ISP back end or a real Sun workstation.

Maybe if Oracle hadn't bought Sun, but OTOH, with the collapse of the non-PC workstation market and inability of Sun to re-invent themselves like IBM (services) or Apple (iPod and then iPhone saved them) perhaps Solaris would have bit the dust ages ago. Where are Silicon Graphics, Apollo Computer, DEC (Digital) today? Or even Wang, Nixdorf, AST, ACT and Compaq?

Still, a pity someone else didn't buy Java and/or Solaris. Mind you I struggle to think of anyone "nice" left from the old days (before 1993).

[tom dial]

Re: Not going to happen

Re: "IBM has gone that route with their newer line of mainframes."

The DoD agency I was with began using Linux on zSeries in 2009, not all that long after IBM introduced the IFL"engine" at greatly reduced rates. Migration of the first application was nearly painless. The only significant issue going from Oracle and HP-UX on PA-RISC to Oracle and SuSE on the z9 was an Oracle external procedure C language coding error that caused a problem in the 64 bit environment. The new environment was superior to the previous HP host system in both reliability and performance. Our only grief was the larger than expected billings from DISA, which took a few years to adjust the billing rate.

[FrankAlphaXII]

Re: Not going to happen

Thanks for that. I now have an image of Larry pissing in a well scorched into my brain that I don't have enough 100 Proof Mindbleach or 22% THC "Fuck-It" to get rid of.

[Paul Crawford]

Re: Not going to happen

Oracle lawyers - yes, they might/probably will stop this

Red Hat - who cares? They are not "Linux" even though they are the biggest commercial outfit.

To add to the tale our work bout a ZFS appliance from Sun just before they were bough by Oracle and it was a disaster, but mostly not for the underlying ZFS system. The majority of stupid problems came from a combination of the Frankenstein "modified, not quite Solaris" version of the OS, the appallingly bad appliance management software, and the fail-over cluster system that mostly would not fail over except for a kernel fault (so it could be locked up, not serving files, but it STILL did not fail over). It had great promise, and feature-wise it was excellent, but the system management daemon was fragile and unresponsive (much more so if there was actually a fault) and simply not fit for production.

[Anonymous Coward]

Re: Not going to happen

simply not fit for production.

ZFS is a nice file system, but the ZFS appliance was designed by a bunch of people who knew how servers worked but had no clue about storage systems, hence the problems you describe. They knew how a storage appliance should look to a client but did not really understand how it worked internally. Compare a ZFS appliance to Hitachi or EMC boxes & it's primitive and lacking many expected features (synchronous replication, split-site virtual LUNs, decent storage management). They also had the historical "Not Invented Here" Solaris mindset so they (re)invented primitive clustering software instead of using the available Solaris solution (which the tape management guys use successfully in their appliances), and as usual with people who think HA is easy they came up with something which is OK when everything works as expected, but is not robust once faults appear. If you want decent Oracle storage, use the Pillar-based FS arrays.

[Anonymous Coward]

Re: Not going to happen

"Red Hat hate Oracle"

there are people that do not hate Oracle?!

You mean that Oracle: "Oracle poisons or pisses in every well it comes across." ?

"they will never allow it to go upstream"

Red Hat is not Linux, and they definitely don't have a veto power

"and would likely exclude it if Oracle got it in upstream"

Red Hat does what's legal and what customers ask for. If it's legal to ship ZFS with Linux, they'll ship it.

[ivant]

Re: Not going to happen

Maybee it will!

[PNGuinn]

Re: Not going to happen

Ever tried p*ssing upstream?

No? thought not.

[John Brown (no body)]

Re: Not going to happen

"Ever tried p*ssing upstream?"

Apparently, you get a nice warm feeling.

[Denarius]

time to buy shares in high grade memory fab then

turn on deduplication and watch the RAM get sucked away. Not that I dislike ZFS, one just has to ensure adequate resources are on hardware.. Given how RAM prices are good from 10 years ago it is not physically improbable but I agree with my esteemed commentard above. The lawyers will make it poisonous to do. Back to the BSD migrations off Linux then...

Query, whats happening to Reiser FS these days ? It did seem better for OLT with lots of small frequent transactions.

[HCV]

Re: time to buy shares in high grade memory fab then

"Query, whats happening to Reiser FS these days ? It did seem better for OLT with lots of small frequent transactions."

It turned out to be murder to keep development going.

[Korev]

Re: time to buy shares in high grade memory fab then

In very poor taste... But I won't pretend I didn't laugh ->

[Mage]

Re: In very poor taste

Ah, but this is the closest to pre-September Usenet. I try not to be tasteless, but if I was put off by some being tasteless, I'd have left years ago.

Probably 23 Nov 2007 was when I changed my user name from Madge, joined 7 Sep 2007, but only 20 posts (user 9464), so I probably had an earlier name.

[Anonymous Coward]

Re: time to buy shares in high grade memory fab then

Makes me think Pottering should have married Hans Reiser...

[Alan Brown]

Re: time to buy shares in high grade memory fab then

"turn on deduplication and watch the RAM get sucked away."

That happens in any large system where you enable deuplication. The short answer is don't do that unless you know what you're doing and the benefits outweigh the costs (they usually don't unless it's a mail spool)

[grinder]

Re: time to buy shares in high grade memory fab then

Not really. NetApp does (inline) de-dup and compression without having massive amounts of memory in their storage controllers like a ZS3-2 and/or a ZS4-4. It was only last year that they introduced beasts like the FAS9000/AFF A700.

Don't get me wrong, ZFS is a brilliant file system and I 'd really like to be GPLd or dual licensed to keep up with the BSD folks. But the ZS appliances were terrible. I faced similar problems to the ones other guys mentioned before me. And it's not only those problems, but also the extremely bad support from Oracle. I had tickets opened with them for literally months. Every time chasing their support guys and account managers to move things forward.

After some point it was obvious that Oracle was not interested anymore in the platform. Fortunately, we migrated away from them couple of weeks before announcing their decisions about the ZS appliances and Solaris without even knowing that they were going to kill them. I am a much more happy storage admin now :)

[bazza]

GPL2? Think of the *BSDs!

Unless the code is released under GPL2, it cannot be integrated into the Linux source code. There's been enough fuss already about using it as a separately provided kernel module.

The problem I see with that is that once it's released under GPL2, will it continue to be released under the more permissive license that helps out, for example, FreeBSD? It would be a real pity if ZFS went to GPL2, and GPL2 alone, because it would seriously screw things up for the people who are already using it elsewhere.

It could be multi-licensed of course, but license fragmentation can easily lead to source code fragmentation too, unless absolutely every contributor is commited to releasing their efforts under multiple licenses.

[DainB]

Re: GPL2? Think of the *BSDs!

And what would prevent Oracle from adding it to UEK ?

[Anonymous Coward]

Re: GPL2? Think of the *BSDs!

"And what would prevent Oracle from adding it to UEK ?"

For the root filesystem they would need to modify the installer and other utilities that would take it a long way off the Red Hat path.

Personally, I wish ZFS was the root and default filesystems. The current methods of increasing the size of a filesystem are rather primitive.

[chasil]

Re: GPL2? Think of the *BSDs!

Many, many people hold copyright on the Linux kernel. They could sue when Oracle violated their copyrights with sections 3.1 and 3.4 of the CDDL:

https://sfconservancy.org/blog/2016/feb/25/zfs-and-linux/

Oracle cannot continue to use Linux/ZFS unless the CDDL terms are relaxed. Sun designed the CDDL specifically to prevent it from spreading into Linux.

Here are the relevant sections.

[§]3.1 … Any Covered Software that You distribute or otherwise make available in Executable form must also be made available in Source Code form and that Source Code form must be distributed only under the terms of this License. …

[§] 3.4 … You may not offer or impose any terms on any Covered Software in Source Code form that alters or restricts the applicable version of this License

"We believe Sun was aware when drafting CDDLv1 of the incompatibilities; in fact, our research into its history indicates the GPLv2-incompatibility was Sun's design choice. At the time, Sun's apparent goal was to draw developers away from GNU and Linux development into Solaris. Not only did Sun not want code from GNU and Linux in Solaris, more importantly, Sun did not want technological advantages from Solaris' kernel to appear in Linux."

[Suricou Raven]

How about fixing btrfs?

How long have we been waiting for them to fix the dangerously unreliable RAID5-a-like mode now?

[Lee D]

Like everything...

Call me when it's in there and it works and I can use it. Until then, it's nothing more than hyperbole.

There was nothing stopping this happening 5 years ago, there's nothing stopping it happening now. Even as a GPLv2-licensed kernel fork, without going through proper submission, it would slowly creep across if it was that useful.

Fact is, though I'd have trusted Sun to do so, and to bring this across, Oracle more often than not ends up as a curator of someone else's code that they deemed destructive to their business so they bought them up. Everything they touch gets forked and the fork is - without exception - better. I can't believe that ZFS is any different. They'll dump it on someone when they no longer can be bothered with it, which is what it sounds like they are heading towards. Maybe some ZFSv2 Linux spin-off will actually make it into the kernel, with some kind of better performance, ideas, avoiding patents, etc. but that's decades away.

And the anecdote about being told "use that broken stuff, no you can't go back despite having myriad problems"? Way to run a business? I don't think so. That doesn't inspire me with confidence at all. The story should be "the person came in and insisted we move to ZFS - against other's people's wishes because it was 'too early' - because they'd secretly trialled it alongside their existing system and it performed better than anything else". Not "It was a heap of junk but we're gonna make you use it and we'll just throw you a few developers so your live-system can act as an alpha test and be patched as we find the bugs that you're constantly running across".

[Roo]

"And the anecdote about being told "use that broken stuff, no you can't go back despite having myriad problems"? Way to run a business?"

That strategy is painfully familiar. The evidence suggests that it doesn't cause enough damage to cause any serious hindrance to multinationals that are 'too big to fail'... Outfits like Oracle can win contracts by simply selling at a loss in to remove the opposition, and/or get a nigh-on-free-cash to buy them out if they choose to. These outfits can't piss money up the wall fast enough to outpace the rate at which pension funds and customers/victims are throwing money at them.

[Mage]

Everything they touch gets forked

Apologies, pasted from Wikipedia:

Hudson was continuing development and that Jenkins was a fork; the Jenkins developers considered Hudson to be the fork.

Interest in Hudson collapsed thereafter. Eventually Oracle donated the remaining Hudson project assets to the Eclipse Foundation at the end of 2012

OpenOffice, is a discontinued open-source office suite. It was an open-sourced version of the earlier StarOffice, which Sun Microsystems acquired in 1999, for internal use. Later they released it as Open Source. In 2011 Oracle Corporation, the then-owner of Sun, announced that it would no longer offer a commercial version of the suite and soon after donated the project to the Apache Foundation. It was forked as LibreOffice in 2010.

What else has had a Forking that Oracle Borged?

[Nolveys]

Re: Everything they touch gets forked

What else has had a Forking that Oracle Borged?

Larry's tongue is forked and he always has that pitch fork with him. Do those count?

[chasil]

Re: Everything they touch gets forked

Solaris. I believe that these are the most well-known forks:

https://www.openindiana.org/

https://www.joyent.com/smartos

I believe that they both use this kernel:

https://wiki.illumos.org/display/illumos/illumos+Home

[Alan Brown]

"They'll dump it on someone when they no longer can be bothered with it,"

Perhaps, but in the meantime the openZFS project is moving along nicely thankyouverymuch and sychronisation amongst the non-oracle OSes is pretty good.

[HmmmYes]

Please, for fuck sake, do.

ZFS and Dtrace almost (almost mind) make up for Oracle releasing Java.

Its worh port the BSD lower file system layer into Linux just to have ZFS.

The only file system me, my kids, and their kids will ever need.

And no mail order brides were murdered during its development ....

[jms222]

May I suggest Debian GNU/kFreeBSD https://www.debian.org/ports/kfreebsd-gnu/ ?

Or just FreeBSD with the Linux emulation layer ?

Or just any other OS that doesn't require you to spend your life worrying about licenses for stuff you already have but might not want to use ?

Jon

[Doctor Syntax]

"May I suggest Debian GNU/kFreeBSD https://www.debian.org/ports/kfreebsd-gnu/ ?"

Unfortunately see also: https://lists.debian.org/debian-devel-announce/2014/11/msg00005.html

"We discussed kfreebsd at length, but are not satisfied that a release with Jessie will be of sufficient quality. We are dropping it as an official release architecture, though we do hope that the porters will be able to make a simultaneous unofficial release."

[Anonymous Coward]

Just use FreeBSD

Simple.

[nijam]

If they'd simply made that oh-so-slight change to the license so that it was GPL-compatible, it would have been both (a) available in Linux very quickly thereafter, and (b) supported in Linux (probably better than it is supported by Oracle).

Not going to happen, though, as other commentards have already pointed out.

[Doctor Syntax]

Is ZFS shipped by Oracle in their Nonstop Linux? If so either Stallman's concerns don't apply or they do in which case Oracle would be obliged to release it under GPL2 anyway.

[Mage]

stability rather than innovation

I love innovation, but if I have to choose, I want STABLE on the production / consumer / whatever system and innovation only on the prototype, in house , the lab TILL it's stable.

A constant "upgrade" cycle of so called innovation that is often tinkering, political or graphic/industrial designer insanity unrelated to user needs and ergonomics isn't innovation. It's change for change's sake!

[Anonymous Coward]

what did they expect?

By 2012, nearly the entire Fishworks team that was responsible for the "ZFS Storage Appliance" had left by one means or another. The people responsible for the core technologies in the product were gone, including both Jeff Bonwick and Matt Ahrens, the original inventors of ZFS itself, all three of the original DTrace team, and all of the 7 Fishworks charter members. Countless other critical, highly talented ZFS and other Solaris engineers had departed, both before and after the Oracle acquisition, for reasons ranging from disgust and frustration to simple mismanagement. In 2011, the head of the Fishworks team was fired outright, and Scott Tracy, a yes man with no technical knowledge of ZFS, minimal understanding of Solaris or the appliance software stack, and a classic middle management addiction to surrounding himself with people who wouldn't threaten or question his authority, placed in command instead.

Given that background and the obviously hostile internal IT attitude toward anything not NetApp (an attitude that is practically universal in the IT world -- if it's not the known, familiar product, it's the enemy), what kind of success would you have predicted? If you want products to be successful, you have to keep people around who understand them. Turning over an entire team is difficult at best and takes years with no guarantee of success. One of Oracle's many failings is its inability to understand that a codebase and patent portfolio aren't worth much without the people who know how they work. Anyone who's ever had any contact with Oracle would have predicted this going badly, and I'm sure many insiders did exactly that. Oracle is a case study in the ultimate limitations of a values-deficient management team.

[Dinsdale247]

Linux people are retar@ds

The Me Toos have it: Linus doesn't like the license so it must be "baaad". "Look," say Linux bashers, " I built an operating system out of spare parts from the GNU project! I can plug anything into it and make my own distro! ZFS? Well unless the filesystems compiled INTO the kernel it's unusable!"

ZFS uptake has NOTHING to do with the license. It's a perception issue. There are "absolutely zero"[1] good resources on ZFS with Linux. The Linux port(s) isn't(aren't) well maintained (or weren't when I was looking), it's WAY behind the other implementations in terms of features and it's buggy. Why won't anyone use it? Because nobody uses it.

Are these instructions to get it working SO onerous that Linux users can't install it? Please...

https://github.com/zfsonlinux/zfs/wiki/Debian

[chasil]

Re: Linux people are retar@ds

As I understand it, CDDL-licensed code requires anything that links with it to also be CDDL-licensed. Shipping a compiled kernel that includes ZFS binary modules would apply that CDDL license to all the other kernel code.

This would open the distributor to a lawsuit from all of the other contributors who did not agree to relicense their GPLv2 contributions under the CDDL.

Here is a discussion of these points (mentioned specifically as points 3.1 and 3.4):

https://sfconservancy.org/blog/2016/feb/25/zfs-and-linux/

This relicensing liability does not fall on a distributor if ZFS is obtained in source file format (*.c) and the user invokes the compiler during the installation (via "dkml"). I know of one Linux distribution that does exactly this.

[FullGear]

Re: Linux people are retar@ds

CDDL code doesn't care what it's linked with... you just need to publish any changes you make. The only reason ZFS isn't part of Linux is that the GPL doesn't allow distributing Linux that includes code under other licenses. The only "restriction" on ZFS is that you cannot relicense the CDDL code under the GPL... same as you cannot relicense GPL code under the CDDL.

[phuzz]

ZFS on Linux? Pffft

Who cares if ZFS will be included in linux, when there's more important news out there: OpenZFS on Windows!

In case anyone feels left out, it seems that the author uses a Mac to develop on, which I think makes them a heretic in pretty much anyone's eyes :)

[Anonymous Coward]

ZFS != ZFS Storage Appliance

I don't think it's fair to blame ZFS for any issues with the ZFS Storage Appliance or whatever Oracle is calling it these days. It's a great filesystem that makes maintaining a Solaris 11 system rather painless. They just really dropped the ball when it came to making a storage device out of it.