So we are now nested, what, six deep in forks? I've long since lost track: OpenSolaris => OpenIndiana => Illumos => this new thing ? are there some in-between steps?
And we don't have 100% solaris api/abi compatibility.
Or a Sun compiler.
Or Sun drivers, necessarily.
Or a real guarantee of future source releases from Oracle.
I don't believe these are all forks. Illumos is a fork of OpenSolaris. OpenIndiana was NOT a fork though, instead it was a distribution of first OpenSolaris and now Illumos.
I compare OpenSolaris/Illumos to the Linux kernel. You typically don't just install them just like you typically don't install Linux. Instead you install a Illumos distribution or you install a Linux distribution. Obviously this isn't a perfect comparison since Illumos contains a lot more of the user land than Linux does.
Current distributions of Illumos appear to include: OpenIndiana, SmartOS, Nexenta, and now OmniOS.
Where do you get that we don't have 100% solaris api/abi compatibility?
One of the pressing projects appears to be to make the Sun Compiler optional. Illumos' preferred compiler to be built with is GCC 3.4.3. They do still rely on dmake and the lint from SunStudio though. Hopefully that can be resolved.
I don't know why we need Sun drivers specifically, unless we are using Sun hardware.
I think we can safely assume now that there will be NO future source releases from Oracle.
As to who this is for, I believe other comments do a good job expressing that.
OpenSolaris has been dead for two years. illumos is a fork of the source code repo that made up the OS and Net consolidations from Sun. Oracle closed the gate shortly after the acquisition, with the promise they'd drop code after Solaris 11 was released. This has not happened.
OpenIndiana, SmartOS and OmniOS are all distributions of the illumos kernel and associated libraries and programs.
OpenIndiana is geared more towards desktop users (though it can be used for servers well enough). OI is funded primarily through everycity.co.uk.
SmartOS is a PXE/USB booting server operating system, where all services are meant to be encapsulated in zones and KVM. SmartOS zones use pkgsrc for packaging (via pkgin). SmartOS is useful for building clouds, for instance. It is the core of Joyent's SmartDatacenter product.
OmniOS is a more traditional server operating system in that you install bits on metal, use IPS for packages, and have an upgrade path other. OmniOS has been used in production inside OmniTI for a while (as I understand it), but is the newest player on the block.
As with Linux distributions, each distribution above has different operating and development philosophies, but they all push code back up through the illumos gate.
Sun is dead. Oracle ate Sun, took their ball, and went home. illumos is the successor of the OpenSolaris idea (sans asinine governance model), and the core of open source Solarish distributions.
People who want to run ZFS for storage usage. ZFS is still better than Linux alternatives like LVM, BTRFS etc. To be picky, there is not an exact counterpart to ZFS under Linux.
And with clones etc. you can set up 1 master KVM image and then clone it 10 or 100 times, saving a lot of disk space in the process (ZFS clones only use the amount of space that is different from the original).
People who have a lot of Solaris experience or have to maintain a lot of Solaris systems already. Also people who are using Solaris zones.
People who are building backend systems that require or can use any/all of the above. Right now, no one cares what your Web SaaS service runs on, as long as it stays up and doesn't lose their data.
Solaris still has a better VM subsystem, IMHO, than Linux does, and performs very well under heavy loads that include disk activity.
Regarding zfs: the zfs on linux project (http://zfsonlinux.org/) is actually fairly mature and has decent performance ( though there are still some outstanding issues with using zvols as xen filesystems, and also if using a xen paravirt kernel ).
You've explained who Solaris is for, but if you want Solaris why would you want to run a fork of a fork of Solaris? It's still not clear what value this adds relative to Illumos.
For one, KVM virtualization does not run on Illumos, nor Solaris.
Second, Oracle is encumbering Solaris with a not-nice license. I haven't followed the latest developments because I no longer care, since I will never use a non-free OS.
Whether you meant to ask, "Why Illumos over Solaris" or "Why SmartOS or OmniOS over Illumos?" I think I have answered your question :-)
illumos is free and (most importantly) open source. Running a closed OS and binaries on a production server isn't the most foolish idea I've heard, but it's in the running.
Most things work on AMD64. In SmartOS, KVM was the major feature that didn't work on AMD64, but as they said in the faq, the community has been working on that. I can't currently find the testing ISO builds with that support, but here is at least one repo with KVM on AMD support: https://github.com/jclulow/illumos-kvm
Joyent's original work is worse than just Intel only, it requires features found in Core iX and relate Xeon CPUs. This means it doesn't work on Xeon 5xxx chips that are getting affordable and it doesn't work on C2D chips that support VT-x. It sounds like AMD support is further along than supporting older Intel chips.
That's my repo! The "pre-ept" branch contains experimental support for both AMD CPUs with AMD-V support, and Intel CPUs lacking the EPT feature (Joyent targeted EPT support with their initial port.)
I have a hacked up SmartOS ISO from a month ago with a build of the "pre-ept" KVM module replacing the stock module. If you'd like to test it out, with the disclaimer that it might not work for you, it's here: http://alpha.sysmgr.org/smartos-20120223-jmc2.iso
If it works for you I'd love to hear about it. If it doesn't work and you can provide crash dumps or even remote console access to the machine for me to debug the problem then I'll try and help out. My goal is to get the new support tested and working on as many different bits of hardware as possible and then get it integrated into Joyent's repo once cleaned up and proven.
A lot of us hang out in #illumos on freenode (IRC) if you'd like to chat -- I'm LeftWing.
EPT (extended page tables) which are required by the Joyent work made it that much simpler to port the KVM stuff over. EPT simply a lot of the work required and reduce overhead for virtual machines.
Given the size of the community, it's likely that Illumos will support less and less hardware over time. You may want to stick to whatever's on the Nexenta compatibility list.
The SmartOS ISO I downloaded, did not have an installer to help you install the OS to disk.
This one does, according to release notes.
Looks like they have tweaked some of the libraries and they provide both 32 and 64bit versions of Python also (this can be a pain to build sometimes as the configure tools don't always pull in the right libs, so you end up with mixed 32 and 64bit objects or other weirdness).
Also, the installer is minimal, basically just asks you what disk to install on. You do all other configuration after you have booted it for the first time.
Can anyone elaborate: why all these cloud webhosting people are taking legal risks of Oracle changing license for (Open)Solaris or closing this OS project entirely, legal risks of being sued by Oracle for patent infrigement / whatever else legal bullshit?
They could just port KVM to FreeBSD instead; FreeBSD already has ZFS, DTrace and all other necessary bells and whistles for cloud webhosting infrastructure. But in contrast to (Open)Solaris they have community support for new hardware, a lot of already working drivers for existing hardware and also much safer legal position.
Are there any technical difficulties with porting KVM to FreeBSD? Can anyone elaborate?
As I said above, OpenSolaris is dead. illumos is a fork. We don't take code from Oracle, because there isn't any. If they'd kept the gate open, a fork probably wouldn't have happened.
As for why not to just move from Solaris to FreeBSD... because they're totally different beasts? And we happen to like Solaris? :-)
Porting KVM to Solaris was apparently quite non-trivial:
The opposite actually, the newer KVM stuff for example will not run on older processors, Intel Core iX and very new model Xeon's are required with EPT support.
Many people, my self included, are using Solaris mainly for big storage where the safety of data comes first and foremost. There is still not good competitor to ZFS that works as well and provides the same gaurantees.
Honestly, more and more people with big data problems are turning back to solaris kernels over linux because of how many issues there are with linux when you are really running it hard. ZFS, dtrace, and the Solaris kernel on new hardware is really a compelling argument over Linux.
I hate to be the one asking for a citation but can you share some examples of "more people" who have moved from Linux back to Solaris? I agree about excellence of Dtrace and ZFS but I'm just curious about your kernel comment.
And we don't have 100% solaris api/abi compatibility.
Or a Sun compiler.
Or Sun drivers, necessarily.
Or a real guarantee of future source releases from Oracle.
Who/what is this for?