Hacker News new | comments | show | ask | jobs | submit login
OpenSolaris governing board threatens dissolution (h-online.com)
80 points by mattyb on July 13, 2010 | hide | past | web | favorite | 62 comments

(Open)Solaris died a few years ago. It has got some great features like Zones and ZFS, and it's still very stable. However, they lost the hearts and minds of almost everyone. I used OpenSolaris for a few months as my desktop. They don't have the package support of Linux, or the large community. Gotta remember to get GNU Make before Sun's Make in PATH, Gnu sed before Sun's, or was the other way around, Perl CPAN modules not compiling... What a waste of time.

As Scott McNealy said: you've gotta get all your wood behind one arrow. That arrow is now Linux.

Solx86 was always the red-headed stepchild. I don't think Red Hat would have gotten nearly the traction in did if say Oracle on Solaris on Intel/AMD was as well-supported as Oracle on SPARC. Classic case of "innovator's dilemma" - Sun were unwilling to risk the lower end of their SPARC business.

I personally have been involved in projects where we've replaced a million dollars worth of SPARC kit with 50 or 100 grands worth of Intel running Red Hat. It would have been easier, cheaper and quicker to have just swapped the hardware and recompiled... TCO is not (just) in the hardware, it's in the people you need to program and operate it. We had guys with 20 years BSD/SunOS/Solaris experience, developers and sysadmins.

But who'd risk their mission critical systems on an OS the vendor is embarrassed of?

I went to an OpenSolaris users group last year. They were talking about new network features like being able to assign sub-interfaces to different zones. Stuff that VMware ESX had available for years. I was thoroughly unimpressed. The concept of Zones or FreeBSD jails has some appealing aspects; not needing to load multiple kernels in memory, for example, but by this time, Hypervisors such as VMware that deduplicate memory blocks pretty much negate any advantage Zones/Jails once had.

ZFS definitely has some potential though. I hope it will be released under a more liberal license that will allow a direct Linux kernel module as opposed to fuse support.

What kind of ratios are you actually seeing with VMware's deduplicate memory features? I have not seen much on anything I am running, though I admit I don't have them overcommitted very much on RAM (total RAM allocated to all VMs is about the same as physical RAM in the system).

This really depends on what you are running. If you're running an eclectic mix of Windows, Linux, all running different applications and software, you might not get much benefit. Most of our VMs are all running Windows Server 2003 (ugh!) and the exact same terrible web application so we get pretty good savings; about 50%.

I went to an OpenSolaris users group last year. They were talking about new network features like being able to assign sub-interfaces to different zones. Stuff that VMware ESX had available for years.

Zones and ESX are very different beasts, and de-duplication of memory is only part of the issue when you're talking about virtualization of user-space versus hardware virtualization.

But why now? There must have been a good reason for Sun to continue developing their own UNIX variant. I can understand the value-add coming from ZFS, but why having to maintain your own kernel? They could have built their own Linux/BSD distribution with ZFS and any other Sun-branded additions. I guess my question is not "why now" but rather "why not 5 years ago".

I don't really know much about Solaris, just trying to understand the business reasons behind their decision to compete with Linux kernel.

Hm... perhaps because they have thousands of paying corporate customers who started using Solaris before Linux even existed, and they needed to transition them to a newer/expensive hardware without affecting their Solaris-tuned software?

(just thinking out loud, can there be any other reason?)

Even now, Solaris has a far superior VM subsystem and is better performing with multi-thread operations (IMHO).

However my company is going to move away from Solaris to Linux because Oracle is simply not trustworthy (I won't say "IMHO" because it is pretty much a fact).

I've been following this for a while. OpenSolaris, VirtualBox and MySQL are all things I was worried about surviving in the wake of Oracle's takeover. I use all three quite a bit.

I was not aware VirtualBox was controlled by Oracle. I just switched to it because it was a free alternative to Parallels. Suddenly, I care about this issue!

I switched from Parallels when Parallels started sucking AND THEN wanted more money for me to upgrade to 3.x. VirtualBox, back then, was anemic but functional. Now, it's really going places with features that still trail behind the commercial brands you know and love, but when you consider it's free, it's hard to complain too much.

I don't think you have anything to worry about virtualbox, virtualization and cloud computing are now instrumental to growth. Oracle did not have a comprehensive strategy. With VirtualBox it gives them a jump start on building a roadmap to one. The other two, yeah they are in trouble. I was worried about Netbeans and Glassfish before the acquisition because it is the most rapid way to develop in Java and the best solution out there in the latest class of IDE / server run-time combos. They just released new versions so I am starting to feel some relief on that one. I got bit on OpenSolaris and OpenSSO I used both of them extensively and they killed both pretty quickly.

I don't think MySQL is in trouble. The MySQL community is strong, and innovation from the likes of Google, Percona, Facebook, and others has been just as important as of late than anything that Sun/Oracle/Innobase has done. There are plenty of ways that MySQL can succeed even if Oracle were to completely abandon it.

I don't worry about MySQL at all.

Add client-side Java and JavaFX to that list. Sun was already failing on the desktop, but Oracle seems to have even less interest. Thank goodness HTML5+js will soon reach enough penetration to replace Java as a portable application framework.

I installed FreeBSD a couple weeks back just so I could play around with ports and ZFS. Getting ZFS to work with the bootloader is kind of a pain (should be fixed in FreeBSD-9, I'm hoping) but otherwise I was blown away by just how intuitive ZFS is to use and how ridiculously powerful. ZFS was introduced in 2005(!) and I'm still amazed by how far ahead of the pack it is. It's really a damn shame that we can't get ZFS into the Linux kernel because of licensing.

if you feel like playing around with filesystems, try dragonflybsd (http://www.dragonflybsd.org/). it has some neat things like its unique hammer filesystem and virtual kernels.

Here's a view from a prominent member of the (rather small) OpenSolaris community: http://www.cuddletech.com/blog/pivot/entry.php?id=1134

He's not exactly impressed.

Remember the OGC is headed by Schilly. Who's quite famous for actively campaigning to stop people patching his OSS cdrecord to use regular device files, and has a reputation of being very outspoken against anyone who disagrees with him.

If the OGC head is a problem, this, certainly, solves it.

Oracle can easily make a new OGC the following day and name whoever they are comfortable with (or whoever they think will help OpenSolaris get developed in the right direction.

This article got submitted too:


Out of curiosity, does anyone here use OpenSolaris for anything?

I use Nexenta, which uses the Solaris kernel, but with a Debian server-style userland, as a NAS. If ZFS were on Linux, I too would switch to it. Straight Solaris is like travelling back in time in comparison to the usual GNU userland, and while it's got nicely integrated ideas - like the services, logging and monitoring, etc., - the cost of being in a less mainstream branch is reasonably high.

At least with Nexenta, you have an apt-get which might work for some common packages. With straight Solaris, I have to go through all my shell scripts and only use POSIX options, etc., which turns out to be quite a pain.

Wait a year or two - RHEL 6 has btrfs (copy on write with btrees) in tech preview now.

Why wait? FreeBSD has support for ZFS now as well as a great package system.

Well for one, Btrfs is more efficient in some cases, according to the ZFS developers.

Don't even wait, Next3 is stable. Out of efficient snapshots, other ZFS goodies are easily covered with lvm, etc.

btrfs is also an Oracle project.

We use Solaris 10. Which isn't quite the same thing, of course.

The Solaris kernel is excellent. The Solaris 10 userland, not so (except for the newer core stuff like SMF), and OpenSolaris is supposedly a Solaris kernel with a saner userland. But I'm yet to be convinced it is stable enough to run in production, and the new packaging system looks mildly troublesome.

And it's still difficult to get it to install without X.

Isn't SMF just parallelism and horrible XML config files?

Hard to figure out whether your service state modification (say, a service restart) worked too. It always returns 0, never has output, and you have to dig for a log file to figure out what happened.

After using svcadm, use svcs (or functions in libscf) to find out what's going on.

And dependencies, configurable restarting based on failures in other services, delegation of administrative privileges to other users, configuration of managed services through arbitrary properties, integration into OS fault reporting, reliable management of server processes using Solaris Contracts.

You're right, the XML config is a bit nasty.

I use it on a few workstations. It's a fine operating system, but I'm admittedly a bit of an enterprise UNIX head that's disenchanted with where Linux has gone. For the last decade or so, I've come to like some of the other free operating systems such as the BSD family and OpenSolaris.

I do, our webservers run it. ZFS with automatic snapshots has saved my ass quite a few times already. Sending incremental snapshots (for backups) to offsite location (another OpenSolaris machine) is also easy and efficient.

But, if linux had ZFS i'd probably use that instead.

FreeBSD has ZFS too. We've been seeing a lot of people migrating from OpenSolaris over the past few months due to Oracle.

I wonder if the OpenSolaris people would consider a merge? Ie, work on zones, any useful Solaris management tools, etc...

FreeBSD already has jails, so there's no point bringing in Solaris zones; but we're generally quick to bring in features from OpenSolaris. DTrace, for example, is available for the FreeBSD kernel now, and work is underway to make it available for FreeBSD userland too.

Could you clear something up for me? If ZFS is so highly coveted (which I've inferred from recent discussions), why doesn't every operating system under the sun support it?

In a nutshell, the issue is that the Linux kernel which is licensed under the GNU General Public License is incompatible with ZFS which is licensed under the Sun CDDL. While both the GPL and CDDL are open source licenses their terms are such that it is impossible to simultaneously satisfy both licenses. This means that a single derived work of the Linux kernel and ZFS cannot be legally distributed.


FreeBSD has it, but it's been largely abandoned on Mac OS X.

Maybe its for the best. I'm not a big fan of autotools, but proliferation of semi-compatible UNIXes was the reason behind this mess. I'd be absolutely happy with just Linux/BSD/Macs to worry about.

I'd love to have a lively discussion here about OpenSolaris, could not upvote this enough (actually I don't think my upvote counts at all, is that because I don't have karma yet?)

Yes, otherwise people would be able to create dummy accounts and vote themselves up for nothing.

Some things I have been observing can be explained by that. I believe this gotcha is being actively circumvented.

Why do they need a blessing from Oracle though? "If you don't like it - fork it" doesn't apply here? Licensing issues?

99.99999% of code contributions come from Sun^H^H^HOracle. What's the point of forking? It'd be much the same outcome, a code base which never quite gets released.

Exactly. The Sun Community License.

I'm not sure, but I'm guessing that 'Solaris' is a trademarked term, owned by Sun (now Oracle). If you fork it, you can't call it 'Solaris' anymore. You'd have a hard time selling/promoting this new unix os that no-one has heard of.

You're right, it would have to be a different name.

However, if there were a sufficient community around it, this would not be a problem. But because it's so driven by Sun, and they never really managed to build an active community, there is no one who would naturally start using the forked OS.

The community around it is not sufficient. The license prevents borrowing code from Linux. It can, however, borrow BSD code.

But not even that would be enough to keep it alive and evolving.

I'm sure I'll get knocked for this (again), but if you want to pick nits, it's the GPL that prevents the commingling, not the CDDL.

Why not just say that the licenses are incompatible?

That's the simple truth, and that's not a slight on either of them.

The real problem here is that there doesn't appear to be an OpenSolaris development community outside of Sun^WOracle. It's much like Mozilla in its early days, though with an even smaller chance of success. Mozilla's competitive landscape was devoid of credible open-source alternatives - that's not the case for OpenSolaris.

Fair enough. Linux also can't borrow code from OpenSolaris. I would love to have ZFS on Linux.

Anyway, it was Sun who raised the possibility of licensing OpenSolaris under GPLv3 (that would still prevent Sun from borrowing lots of Linux code and Linux from borrowing any Sun code, but it would, at least, be a start) not the Linux folks who considered re-licensing Linux under a CDDL-friendly license, hence the "unidirectional" slant of my post.

As for the community and Sun^WOracle's contributions, Oracle has made it clear that they would follow an open-core model for the OS.

That's really sad, IMHO, but predictable. Ellison is nicer than Gates and way cooler, but it doesn't make him a particularly nice guy who wants to share.

BTW, I think the copyright fragmentation you see in Linux is one of its strengths. When you contribute a patch to OpenSolaris, do you retain its copyright or you transfer it to whoever controls OpenSolaris?

That's fair - Sun at least had the option of relicensing (given that the required contributors to sign an IP agreement).

FWIW, it saddens me to watch Solaris stumbling towards its grave. Remaining available only on hardware sold by Oracle renders it worthless to me.

I can't make that shift now, which means that I need to build my own credible alternatives to using Solaris. And when I am in a position where I could afford Solaris on Oracle hardware, I won't be so inclined - because I'll have my homebrew alternatives, and a bitter taste in my mouth.

So it goes.

May I be the first to suggest calling it either "Rheya" or "Kelvin"?

Some of it is still closed source (mainly the libc, if I recall). You lose out on patching those parts if you don't play nice with Oracle.

That's a threat?!

Larry Ellison: "Uh, sure, we'll get right back to you on that."

[Larry alt-tabs back to his word processor with a draft of his Solaris Developer Trial licensing plan: foree for 30 days followed by a $9000-per-seat licensing fee]

Larry may not be the nicest guy around, but he's no idiot. He just spent a ton of money on Sun and something like this would put that much more momentum behind commodities he cannot control (like Linux)

Indeed, Ellison is no idiot, and that's precisely what he's been doing: putting Sun's formerly free properties behind strict paywalls in order to get the company cash-flow-positive again after Sun bled money giving away the geese that laid the golden eggs.

They only laid golden eggs because they were relevant. If his clients decide to migrate away from Solaris, they may also decide to migrate away from Oracle. Being legacy-only is not good for long-term revenue.

The Imperial Senate will no longer be of any concern to us.

It is now as dead as say Digital Unix or OpenVMS.

Applications are open for YC Summer 2018

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact