Inside Mac OS X 10.7 Lion Server: Apple replaces Samba for Windows networking services - Page 2

No, actually we (the Samba Team) are quite cool with it. Apple makes their own commercial decisions to decide what's best for them. If they want to move away from GPL software, that is their right and we have no grounds for complaint.

I think they're wasting resources, as we (the Samba Team) will maintain a SMB/CIFS/SMB2/AD server for them if they want, but if they find our license unacceptable then they need to do what's right for their users.

We will still try and produce binaries for Samba 3.6.x and Samba 4.0.x (which will include full AD support) for users who wish to continue using Samba on OSX.

Jeremy.

Hate to drag you back into the real world, but IBM, Netgear, CISCO, Google, Symantec and other are regular contributors to GPLv3 Samba, and have products based on it, or using it. It's a shame Apple doesn't feel confident enough to do the same (especially since they employ a Samba Team member with direct commit access to our repository) but that's their call.

Cheers,

Jeremy.

Terrible news.

Things are really looking down for OS X these days.

Yet many do anyway.


If wasting resources was a concern then Samba staying at GPL v2 or later would have been a better choice. GPL v3's patent provisions are more business hostile as are the DRM provision, hence the avoidance by many companies.

You pushed this as a screw move to Novell over the Microsoft deal for ideological reasons as opposed to practical ones.

This is one major reason that Linus > Freetards.

Amusingly, GPL usage as a percentage of OSS projects has been on a steady decline since GPL v3. 70% in 2008, 65% in 2009 and 61% in 2010.

http://www.blackducksoftware.com/new...ses/2009-06-30

http://www.blackducksoftware.com/oss/licenses

So on the plus side there are more permissive licensed open source alternatives to GPL encumbered ones. Yay for LLVM and Clang!

With any luck Apple will open source their SMB solution under a permissive license as well.

Not to get too deep into NDA territory but if the current dev seed of Lion already has these changes (and I can verify in the AM) then it seriously explains the speed difference. It is noticeably faster. I have a 20TB Xsan system and 10.6 never worked quite right when trying to connect but Lion is super fast. The files are displayed quickly and Finder behaves as I would expect. I attributed it to changes in Finder but it could be more than that.

For all those Corporations contributing to Samba 4 you'd think it wouldn't take 7 to 8 years to reach that milestone. Or am I only assuming Samba 4 is being aided by IBM, Cisco, Google, etc., as Jeremy is implying?

At any rate, I agree that Apple moving in the direction they are makes a helluva a lot more sense than expending resources on a third party solution that is taking eternity to get to their latest milestone. Not to beat a dead horse but Samba4 is nearing their 15th alpha release.

Apple can afford to pay Microsoft what is necessary in licensing and focus on other areas of research and development.

Anyone still running a network on the NT 4 PDC/BDC model should be fired. It may be "simpler", but it's so horribly limited and flawed that it's nearly unusable. Any IT worker that argues for a NT domain should be fired on the spot for not keeping up with the last decade of IT change. Most Windows admins are working on their Windows 2008 R2 migrations. If you're not even up to Windows 2000 technology it's time to get out of the game.

Well, replicating an AD controller is quite hard :-). You may remember a lawsuit in the EU over access to the documentation to allow third parties to do just that..

I don't know, but doubt that the new Apple code includes AD functionality. If it does, well done to them. If not, well, when Samba4 comes out it will still work on OSX so you win both ways.

Jeremy.

Many many NAS, router, and other embedded devices distribute the source. That's the difference. Aside from Darwin, Apple won't go FOSS.

Honestly, ridding OS X of Samba is a welcome change. Samba 3 has been out forever, and still has many of the bugs that have been around since the 1.x days. Performane is horrible (compared to Windows and Solaris) - on just about any platform / hardware. One of the best things Sun has ever done was move to an in-kernel CIFS server. The performance difference is night and day.

One more thing...

People keep referring to Samba and the CIFS client as one in the same. mount_cifs is actually derived from FreeBSD source - not Samba. So mounting remote filesystems has nothing to do with this - it only has to do with SMB SHARING.

I'm talking about Apple wasting resources here, not the Samba Team. We're going to be doing what we do anyway.

I support GPLv3 not as a "screw Novell" move (if you actually read the GPLv3 you'll find that the Novell Microsoft agreement is explicitly allowed, not prohibited), but because I believe it's a better license for Free Software in a dangerous software patent environment.

Using insults like "Freetards" doesn't improve your argument, it weakens it I'm afraid.

I hope that Apple do release their code under a permissive license, it would be good to compare and learn from it, and if it's GPLv3 compatible (as most permissive licenses are) maybe we can incorporate parts of it into Samba. I am encouraged by Apple's recent decision to release their DCE/RPC stack under a permissive license, which I thoroughly praised.

Jeremy.

This isn't all of the pieces, but Apple posted this project on Mac OS Forge at the end of February: http://www.dcerpc.org/

I really like how Apple is unafraid of replacing entire portions of their software stack for implementations they control. It'll be interesting to see if they open source their CIFS client as well.

I know little about the topic. But I wonder aloud:
1. Apple has seemed less than committed to professional users lately. Is this another sign?
2. Apple discontinued it's pro Mac server...is this related?
3. One can only wonder how committed Apple is to mixed networks. They seem increasingly proprietary.
4. Could Apple just be playing hardball/chicken with the Samba people?

I will be needed to invest in a robust mixed network in the next year. I wonder how this will impact us little guys who aren't network gurus.

I'm amused by the comment:

"One of the best things Sun has ever done was move to an in-kernel CIFS server. The performance difference is night and day."

Have you ever actually *tested* the performance ? I'm guessing not. There is a reason IBM's benchmark-winning SONAS product uses Samba you know :-).

Jeremy.

Apple already do open source their CIFS client. It's part of their Darwin releases I believe.

Jeremy.

Of course we're not talking about your not wasting Samba team resources. Just everyone else's.


And of course the it never occurred to you that forcing MS to pseudo distribute GPL v3 code that they wouldn't be able to adhere to their end of the agreement (coupon).


When the shoe fits.


Yes, one way sharing is always good for GPL projects and you would approve.

Nice of you guys to tell BSD devs about MS-DOS seekdir/telldir bug workaround when you guys found the bug. Oh, wait. You didn't. Not your job I suppose and I dunno if they count as upstream, downstream or just not all that important regardless.

Your accusing others of wasting time because you guys opted for GPL is an amusing pattern. I remember you kvetching about userland API access in Solaris and Sun wasting their time (which of course would have been far better spent supporting you) reinventing the wheel with native cifs support in their kernel. I recall you being such a whiny little brat that Alan had to bitchslap you for being such a drama queen on his blog.

Why do folks "reinvent the wheel"? Because the wheel you provide comes with shackles...so much for being "free".

Aside from the argument of GPL for a secound.

What alternative could Apple use? I mean Samba took YEARS to be developed into current state.

I mean Apple cant just invent something out of nothing.........

I don't understand what the all the hissy fit is over gpl3.

Though the samba guy says otherwise, I understood the main reason behind gpl3 was to prevent certain companies aka Novell from doing patent idemification only for themselves, and not for everyone.

Course Linus doesn't care whether or not DRM is used or other commercial uses. His choice as well.


That's like been the case since day 1???

You mean where they blew the doors off the SPECsfs benchmark...for NFS speeds? Where's the benchmarks for CIFS?

http://www.spec.org/sfs2008/results/sfs2008.html

If you have the benches to show that Samba is faster than Solaris CFIS feel free to post them.

Patent indemnification is a big area of differentiation between 2 and 3. I'm not a lawyer, or even a patent lawyer so if you need legal advice should should find some (really). But if you want to get *really* depressed then go to groklaw and read up on the SCO saga that's still going on (amazingly). Then check out yesterday's lawsuits from Microsoft against Android.

I don't have a good answer but even a dumbass like me can see there's a huge battle going on and that means that there's big bucks involved

Am I right in remembering that Apple has had a fair number of serious problems with it's own AFP? So now they are creating their own method to interface with Windows networks? This doesn't sound like it will have a happy ending.

Microsoft keeps changing their side of SMB. Samba then updates itself. Then apple doesn't update its version until much later. Or so anandtech said.

SMB pretty much sucks anyhow. ~25% of throughput is lost to overhead. NFS doesn't have the easy browsing capability though.

Well, Apple doesn't seem too keen on CDDL solutions either but they could go with a port of CFIS and grab ZFS if so inclined. Not likely as they are rumored to be building their own next gen filesystem.

They certainly have the resources to buy and tailor a commercial solution as well or strike a deal with Microsoft. Samba took YEARS because no one was pouring much cash into the development. There's maybe 30-40 commiters for Samba with a core team of a dozen or so. That's like 20 times the size of the Gimp team (2 devs) but it's not all that as I don't think Samba is corporate sponsored with full time corporate devs. Jeremy will correct me if I'm wrong.

Ohloh vastly overestimates the effort and cost for projects. (I have a good feel for how much the ones I worked cost and the Ohloh estimates were insanely high...how? Because I actually knew the staff years that went into them and the salaries of the coders involved.

So using their numbers as the worst case replicating Samba is 371 staff years and $20M. Cost wise that's chump change for Apple and they can buy a commercial implementation that will be 80% of what they want.

"When the shoe fits." - no, it's still rather childish I'm afraid.

Still, I do see that you have some history to recount, as you seem to know Alan. Are you someone I know and have worked with at Connectathon or the SNIA events perhaps ? You don't have to hide behind a pseudonym. I don't. As I recall we had an interesting and lively discussion on Alan's blog. You don't need to resort to words like "bitchslap" to make a point. Correct language is usually enough when you have the facts on your side.

We told the *BSD developers about the telldir() bug many years before it got fixed on that platform. We didn't write such a workaround lightly, and tried for several years to get them to acknowledge it as a problem, and create a fix. Unfortunately it was an Apple employee (who shall remain nameless, as he doesn't work there anymore :-), who was on one of the *BSD core teams, who assured us that what *BSD was completely POSIX compliant, and we would just have to work around it.

Still I'm glad they eventually fixed it :-).

Yes, Sun's efforts would have been better spent working with us on Samba, as I pointed out to them before they even licensed the AS/U code from AT&T. All of their work seems to have given them little traction in the storage space. Most products out there are still based on Samba, and even OpenSolaris OEM's (other than Oracle of course) are looking to move off Sun's (more accurately the code they purchased) rather poorly architectured kernel implementation.

The good thing about the external forks of OpenSolaris are that they remove Sun's rather obstructionist attitude, and allow us to be more creative when working with the external OpenSolaris community. I'm sure you're looking forward to what we come up with :-).

Jeremy.

Unfortunately I'll have to wait until IBM sees fit to publish what they have. Still, I work closely with developers on the Team so I'm rather confident :-).

Jeremy.

There are some sponsored full time corporate devs. About 3-4 working on the Samba3 codebase, and about 2-3 working on the Samba4 AD server. So that's not quite as large as you estimated, but in the right ballpark.

I think you're approximately right on the estimation to reproduce Samba, it's not that hard (at least for SMB2, SMB1 is a little trickier because of all the special cases). The hard part of working on Samba was *discovering* all the implementation details when working with the Microsoft clients. Once that information is out there it's pretty easy to read from the Samba code and replicate using our test suites.

The good reason for using Samba is that most of the core original developers are still working on it. Even Microsoft don't have that (all their devs made too much money and quit years ago :-). Having that history with the protocols and implementations and so being able to fix things quickly is one of the intangibles that makes Samba more useful and valuable to corporations (IMHO of course).

Jeremy.

@jeremy As far as I know we've never met unless it was at a GSOC conference/meetup/whatever. All of my OSS stuff is in a completely different domain.

It's actually really funny, in its own strange way. I was just reading about the latest Debian distro having install problems if you have hardware that doesn't have... wait for it... open source FIRMWARE! Yep, got to find an open source firmware if you want that ethernet / video / whatever to work with their install package. Stallman and the GPLv3 folks are pushing the envelope of killing their own movement.

In terms of their utopian vision, it has essentially backfired, and I assume they're trying to address the now traditional problem with Open Source which is that companies, not end users, end up being the biggest benefactors. Companies have mostly not had to pay for the Open tools they use, but certainly have been profiting from them - and users STILL end up with something proprietary that they can't modify and examine at will. Of course, the idea that most users have any interest in doing so is part of Stallman's bizarre world-view to begin with.

And in some cases companies invest in FOSS - like, say, Apple and WebKit - and then have someone else come along and pick it up, re-package it, and push it as their own (Chrome!). Hey, speaking of companies that could afford to pay... Anyway, I guess I'm just too 'pro-prietary'.

I should have seen it coming too. Apple 'thinks' strategically and dumping XServe seemed isolated.
Smart move.

J.

There are other open source CIFS implementations. Sun bought one and open sourced it; it is in OpenSolaris/OpenIndiana and Solaris 11.

Apple had no problems with the CDDL license attached to dtrace, so should have no problems with the CDDL license on Sun's CIFS code.

Firmware is just another name for what you refer to as a "driver" - they can't legally distribute a GPL product with non-gpl drivers. And if you keep up with any of that stuff, binary "blob" drivers are looked down upon because they usually suck.


You received free, you give free. Chrome is mostly BSD anyhow. If apple or anyone else wanted they can take all the improvements for themselves.

I know that's totally evil for the people here, but whatever.

I think that -even despite the fact that you are so close to the fire as a Samba dev- need to be dragged back into the real world yourself. The fact that some companies contribute and use some GPLv3 code for some things, does not imply GPLv3 is not preventing other companies to do the same for other things.

Example from 'the real world': I've been working as a software engineer at various companies for about 10 years now, all of them who created very specific, industrial software, all related to IC design and production. All software full of highly confidential and/or patented, non-trivial technology that can impossibly be 'shared' or licensed with anyone because it's exacly, you know, what they make their money from. At all these companies even GPLv2 was considered 'toxic' and it was absolutely out of the question that GPLv2 code was used in anything but standalone, low-risk simple front-end stuff. MIT license, BSD license, all of them no problem at all, but commercial companies are allergic to GPL code and will avoid it as much as possible.

Now as I understand (I don't know all the details) GPLv3 is even more restrictive. Hence, you can be 100% sure that this means that by licensing your code under GPLv3, many companies, especially software companies, will avoid your software like the plague. Using it to solve common practical problems is fine, but using it to build your software around: out of the question.

1. In what way is it a sign? And a sign of what, exactly?
2. Goto the Apple online store, click on MacPro, notice how one of them says server? it's actually a new addition. The XServer is also still alive and kicking.
3. The only proprietary networking I see in OS X is afp. Other than that, it has everything else.
4. Apple are very picky with their licenses and they obviously don't like the GPLv3. Plus, with this new, in house tech, they have complete control over it rather than waiting for an open source team to complete it when they feel like it.

This is beneficial in every way I can think of.

Will the new SMB include the DFS client functionality?

Much larger and less easily rackable, but sure, "new addition".


If by 'alive' you mean 'discontinued' and by 'kicking' you mean 'removed from the Apple Store', then yes.

I've been doing this for nearly 30 years, both proprietary and Open Source, so I think I have a wider perspective about this, also from "the real world" :-).

I wasn't talking about all software for all things being GPLv3. There are a range of licenses designed to do a range of things. I happen to think GPLv3 is a great license to collaboratively build an SMB/SMB2 server, and the companies I mentioned agree, and participate. It's a shame Apple don't think the same way, but that's just life.

Hmmm...I thought licensing was why they dumped ZFS. Perhaps it was the potential patent issues but I thought that got resolved.

I'm not saying GPLv3 is bad or that Samba should not use it, just that it would probably see more use and more contributions if it had a license more like the BSD license. That would allow software companies to actively incorporate and contribute to the project instead of just using it as-is and deploying it alongside the stuff they write themselves. The example I gave where software companies often shun GPL licensed code or try to work around it in ways that allows them to use GPL code without 'infecting' their own codebase was just meant as an example to show how GPL licensing *does* in fact limit usefulness for other companies than just Apple.

The fact that GPLv3 asks even more 'freedom' from people using GPLv3 licensed code does not really help here. After everything I've read about GPLv3 I'm still not sure what good it brings compared to GPLv2.

I don't think it was the licensing but rather the lawsuit between Sun and NetApp that scuttled that ship.

Oh I understand that GPL licensing limits the use for some companies, but I still think it was the right choice for Samba. You have to remember the context - back in the early 1990's DOS and NetWare were kings, Windows and SMB were up-coming technologies, but Microsoft (and IBM of course) refused to release the documentation describing how it worked. There were several proprietary versions that were also written without help from Microsoft. The third-party SMB market was fragmented and no one understood all of the proprietary technology.

Samba unified these implementations into one. I think the reason we were able to do this was due to the GPL-share-alike provisions. Had we been permissively licensed the temptation for companies to discover one more "secret" bit of the protocol and keep it to themselves would have been overwhelming. Remember, it's not like Apache where most the protocols they were implementing were open. We're now at the point where it's now significantly easier to reproduce an SMB server - partly due to Samba being FLOSS (and thus readable) and our test suite (which Microsoft now use for Windows), partly due to Microsoft's documentation dump after the EU lawsuit. So we're kind of a victim of our own success here :-). Still, that's not a bad place to be.

The main good GPLv3 brings over GPLv3 is better software patent protection for the project. If you've been following the news recently I hope you see why this is becoming more and more important. Sort of off-topic, but software patents really are a threat to all software engineers, they don't distinguish between open source or proprietary code :-(.

Jeremy.

Apple and Sun haven't said why Apple dropped ZFS. The best rumours were to do with patent liabilities, but also ZFS wasn't very good at handling things like removable drives, that Mac users expected to "Just Work".

As Apple had already put CDDL code (dtrace) in the kernel and system libraries I don't think CDDL was the issue.

An ex-Apple filesystem guy is beta testing a Mac ZFS solution at http://tenscomplement.com/ if you're interested in ZFS. There's also the mac-zfs project at http://code.google.com/p/maczfs/