Hacker News Reader

Comment by xeeeeeeeeeeenu | Hacker News Reader

I wouldn't call illumos "thriving". Commit activity[1] is low for a project of this size (compared to e.g. BSDs) and its hardware support is poor.

Probably the biggest problem is the fact that all existing distributions are undermaintained and unpolished. SmartOS is the only exception, but it's not a replacement for Solaris which was a general-purpose server OS. SmartOS is merely a bare-metal hypervisor.

I really hope that some of the laid-off developers will start contributing to the project. It really needs them.

[1] - https://github.com/illumos/illumos-gate/commits/master

View on HN

While I'm obviously biased, I think you are succumbing to a kind of cargo culting with respect to both commit activity and hardware support.

On commit activity, we in illumos don't necessarily have the same values as other projects: we really want to get it correct the first time -- a trait inherited from Solaris that we call "FCS quality all the time."[1] As such, commits have to be tested, thoroughly reviewed (four or more reviewers is not uncommon), and (as a practical matter) are more often than not shipping in a downstream repo for an extended period of time before they integrate upstream into illumos. In short, we value quality much, much more than quantity -- and what you perceive to be "low" activity I perceive to be quality activity (and as someone who very much relies on the ability to stay current and deploy into production, high commit activity would leave me nervous).

In terms of hardware support, you need to be specific. Yes, there is esoteric hardware we don't support -- but we in illumos are really focused on the server-side, where there is less hardware variation than you might realize. At the moment, the biggest hardware limitations aren't in the traditional domain of CPU/NIC/HBA/flash/spindle (where we have broad support), but rather in the emerging domain of the GPGPU. And anyone who is interested in engaging in the substantial kernel work in this domain (both to make NVIDIA-based GPGPUs to work in hardware virtualized guests and to allow AMD-based GPGPUs to work natively) should consider reaching out -- Joyent is very much hiring![2]

[1] http://dtrace.org/blogs/bmc/2015/09/03/software-immaculate-f...

[2] https://www.joyent.com/about/careers

This is worth pointing out. To me high commit activity is much more an indicator of poorly designed software than a quality project especially when a project has been focused on solving a particular problem for some time.

This Solaris news is disappointing. I feel bad for those of you who have worked on this project for a significant portion of your career. Hope those who have been working on it find a good home somewhere.

Bryan, do you expect this will affect the direction of SmartOS in any significant way?

For us, with the exception of the human toil on former colleagues (which shouldn't be minimized), this is a total positive: it eliminates the Solaris confusion (which we've seen recently in Go[1]), and it gives us a new talent pool to draw into illumos. While I don't know how many will matriculate, I can say with confidence that it will be significant in terms of magnitude if not in terms of numbers. So yes, I expect it to affect the direction of SmartOS -- and very much for the better!

[1] https://github.com/golang/go/issues/20603

Has anyone tried to boot smartos/illumos in Xeon phi knights landing chip (not the coprocessor card). Does it work.

Also how well does intel's userland network and storage libraries dpdk and spdk, work with smartos.

what is FCS? flight control system is the best candidate from reading the link (which doesn't use the acronym) and google.

First customer ship. Historically it refers to a first release of a new Solaris (any Solaris) version. That version needed to pass all known regression tests and needed to be suitable for use in mission critical production environments.

Any time a commit to the code base would be made, it needed to satisfy that criteria, else Jeff Bonwick (the father of ZFS, who was the gatekeeper back then) would unceremoniously rip it out. Any kind of breakage, even during normal development, was unacceptable. This engineering ethos lives on in illumos and SmartOS, which is why commits to the code base are all about quality, not quantity. The commits might not be as frequent as elsewhere, but can pack tremendous amounts of functionality and must pass both peer reviews and all kinds of regression testing, not to mention have a complete and usable documentation, before being committed. In addition to all of this, as far as I am aware, no commits to the illumos code base may be made without a corresponding tracking identifier.

I believe it's "First Customer Shipment" and it's similar to what Microsoft call RTM "Release To Manufacture". You can see it mentioned in passing at the bottom of this page : https://blogs.oracle.com/patch/solaris-10-0508-update-5-patc...

SmartOS is better then Solaris for servers. It's truly badass. I have an HP DL160 G6 running SmartOS in my basement.

If you learn to use zones (which you should) it can run any workload that Solaris can.

On top of that it also runs native Linux containers and KVM for virtualizing everything else.

It's hands down my favorite operating system.

Well, and as of Friday, of course, SmartOS is infinitely better than Solaris.

I can relate to your statement to some extent. Why illumos is not "thriving" is _maybe_ because it does not have solid general purpose distribution. Tribblix is kinda trying to get there, but I haven't used it for more than few sessions inside VM.

Dream OS for me would be something ala OpenBSD, but built on top of illumos and with all the goods that would come with it.

Wouldn't that be OpenIndiana?

Hardware support is by far the stupidest complaint that prevails when people talk about operating systems. Choose your hardware for your software, not the other way around. There are so many better things operating system authors have to do than to implement thousands of competing proprietary hardware interfaces. Mature support for even 50% of consumer hardware would take a massive team of full time engineers years to pull off, and would be wasted when that team could have been making huge strides and innovations in the parts of the OS that actually matter.

If your hardware isn't supported and you don't like it, get your ass into the code and build support for it.

When you say "choose your hardware for your software", you are assuming the software has already been chosen by the time the hardware is selected. That's not necessarily the case. I already have the hardware, it's been bought years ago, and it works with the software I use. Now if for instance I wanted to evaluate OpenSolaris, but it didn't run on the hardware I have, I would probably pass it up. I wouldn't invest hours and hours coding support for it to run on my hardware, before I were sure I would have a use for it. And that's assuming I could; I do happen to know how to write a low-level driver, but many people who would otherwise be interested on the software might not have that knowledge and/or ability.

Of course, once you already use the software, the situation inverts itself. I do choose only hardware which is known to work well with Linux, since that's the operating system I use, and I have even ported a device driver when that wasn't the case. But that can only happen because, many years ago when I first tried Linux, it worked perfectly with the hardware I had.

I think that this argument holds up, but fails to justify the huge engineering cost of supporting large amounts of proprietary consumer hardware.

If you want to evaluate it but don't have the hardware, try qemu.

What kind of domain are you in that allows you to migrate code to random operating systems with such ease? In any case, I would be skeptical if changing OSs mid-business was the common case. More likely, I'd guess people choose the software stack before the hardware, based on their own experience, advice from others and/or other external factors (requirement to run existing applications, etc).

If you wanted to evaluate it, surely you'd use a virtual machine?

Illumos has a lot of nice features I'd like to use. But my product is a small low power ARM device and I'd run out of money long before I'd finish adding minimal ARM support to illumos.

Things are not as simple as you suggest.

Minimal ARM support would probably be pretty easy. 3 or 4 weekends tops.

Hardware support is by far the stupidest complaint that prevails when people talk about operating systems. Choose your hardware for your software, not the other way around.

Amen to that!

> . Choose your hardware for your software, not the other way around.

Oh, look at Mr "I don't need to use what's laying around" over here.

It's a lesson one learned fairly quickly in the late 1980's and early-to-mid 1990's on the Commodore Amiga. Since then, I've always chosen my hardware for my software, because software is infinitely more important than the hardware: even the most powerful hardware is utterly useless without the software to enable one to get work done.

Always choose your hardware for your software, never the other way around.

My point is that you often don't get that luxury.

I believe most effort moved to OpenIndiana.

Reply on news.ycombinator.com