The sincerest form of flattery

[Posted May 15, 2007 by corbet]

Sun Microsystems has made a big show of its open source Solaris release and its attempts to build a working development community around that system. So a number of members of the OpenSolaris community were rather surprised when the press started running articles stating that Sun had decided to embark upon a project to make Solaris look more like Linux. This community was of the opinion that, if it was expected to endorse and participate in "Project Indiana," it might have been nice to know before Sun employees started talking to the media about it.

The person behind this effort, of course, is Ian Murdock, formerly of the Linux community. His position now can be understood from this interview:

When people say they want Linux, they don't actually mean they want Linux. What they want is the Linux userland user environment and the Linux business model. They want choice. They want the Linux distribution and I'm the Linux distribution guy.

Project Indiana, it seems, is Sun's attempt to win over all of those people who only think they want Linux, but who really want a version of Solaris that looks likes Linux.

Many of the goals of this project, as far as they can be determined at this early stage, would seem to make sense. Better package management, for example. More device drivers. Easier installation. A more Linux-like user space with our (relatively) bleeding-edge 1990's shell. And, says Ian, a switch to timed release cycles:

The big feature from my point of view though is the 6 mo. timed release cycle. Timed release cycles have done wonders to introduce predictability into other open source projects (e.g., Gnome, Ubuntu). And 6 mos. is the clear winner in terms of frequency among Linux community/developer distros--it's just enough time to do interesting work AND have a reasonably long hardening period so the thing is stable.

Ubuntu comes up frequently in the discussion; it's clear that some people at Sun see Ubuntu as a model worth emulating.

For those of us who have been working with free software for a while, there is a certain irony in this whole plan. A Linux-like Solaris is not a particularly new concept; for many years, that's how much of the community experienced free software. Before there was a Linux system in a reasonably usable state, the best system to have on one's desk usually came from Sun. As soon as it came in the door, however, it would be loaded up with crucial packages like the X Window System, gcc, netrek, emacs, and so on. Many years ago, we all had systems which, in some ways, looked like what Project Indiana is trying to build now. Those systems did not keep an awful lot of us from jumping to Linux, though, and their cost was only part of the reason for switching.

We switched to Linux because it was free, alive, fun, and clearly going places. There was always something new and interesting happening, especially in those days when running development kernels on production systems was a necessary part of making things work. All these years later, there is still always something new and interesting, and, often, it even comes nicely packaged on a regular schedule. Not many of us are looking back to the systems we used to run.

So it is no surprise that the folks at Sun are putting such a big emphasis on trying to duplicate the things that Linux does right. A similar user space, timely releases, easy upgrades, and, especially, the creation of a vibrant community around Solaris. The thinking seems to be that, if they make a system which looks like Linux but which contains their kernel (which they feel to be superior - a view which is not universally shared in the Linux community), the world will flock to their door.

There have been no real (public) decisions on how this project will proceed; the process for creating an official OpenSolaris project has not yet begun. There has been some initial discussion where it has been suggested that the project start by adopting the work of either BeleniX or Nexenta. This idea drew an immediate complaint from our old friend Jörg Schilling, creator of SchilliX, but it appears that the OpenSolaris community listens to Jörg about as much as the Linux community does. Regardless, it will take some time before the real shape of Project Indiana emerges.

It will take even more time before we see if this project has any real impact. Certainly it should make life easier for Solaris users. But "a better Linux than Linux" is not a particularly compelling sales message. It might just turn out that people who say they want Linux actually want Linux, not another system dressed up in similar clothes. Imitation may be the sincerest form of flattery, but it is usually a poor way to regain one's past prominence.

to post comments

The sincerest form of flattery

Posted May 17, 2007 1:33 UTC (Thu) by kwink81 (guest, #33926) [Link] (5 responses)

We may laugh at Sun now, but things might look a lot different if they switched to the GPL. Many of us believe that the GPL is one of the primary factors of GNU/Linux's success. It has a lot of fan support, and would win Solaris a lot of mindshare if it was employed.

The sincerest form of flattery

Posted May 17, 2007 4:35 UTC (Thu) by zooko (guest, #2589) [Link] (4 responses)

Do you mean, because by switching to the GPL they would enable combination of GPL'ed code and their code? Or is there something else about the GPL that you feel makes it better than the CDDL?

Thanks,

Zooko

The sincerest form of flattery

Posted May 17, 2007 5:00 UTC (Thu) by drag (guest, #31333) [Link] (1 responses)

Licensing compatability mostly. One of the reasons why the CDDL was choosen because it was GPl incompatable. If Sun adopts the GPL then the licensing differences between most of FLOSS and Sun become mostly irrelevent.

It looks like they like the directions the GPLv3 draft has gone. Their CDDL may even be compatable with the GPLv3 anyways...

The sincerest form of flattery

Posted May 17, 2007 17:41 UTC (Thu) by JoeBuck (subscriber, #2330) [Link]

What makes it incompatible is that Sun licenses some of their patents only for CDDL code (implicitly threatening Linux developers who might want to clone dtrace or zfs).

GPL advantages

Posted May 17, 2007 9:20 UTC (Thu) by man_ls (guest, #15091) [Link] (1 responses)

The GPL is the license which best ensures what Linus calls "tit for tat", i.e. it is hard to take proprietary. People tend to cooperate with more interest if they feel their work will remain free.

Besides, most people related to libre software already know the GPL well. Why learn about a new license? The CDDL has only been around for about two years so its effects are not fully understood; economy of effort would suggest sticking to a license known to work.

GPL advantages

Posted May 21, 2007 22:51 UTC (Mon) by salimma (subscriber, #34460) [Link]

One other problem with the CDDL is the stipulation that licensees have to agree to any licensing dispute to be settled in the jurisdiction of the copyright holder's choice. If you plan to use CDDL code that's been partially developed by developers in far-flung countries, the travel cost involved in a licensing dispute alone would be prohibitive.

Choice of kernel

Posted May 17, 2007 7:50 UTC (Thu) by pjm (guest, #2080) [Link]

I think most of what makes Linux

alive, fun, and clearly going places

is actually in the userspace associated with Linux (gnome, kde, ...) rather than Linux itself. And certainly

comes nicely packaged on a regular schedule

isn't affected by whether the kernel is Linux or Solaris: Linux itself has a most irregular schedule.

It's interesting to look at Nexenta and other Debian ports to non-Linux kernels. To an outsider like me (who hasn't used any of them), it appears that they have made substantial success: that large chunks of Debian userspace do work fine on non-Linux kernels, including much software I'd describe as free, fun, alive, and clearly going-places. (Hey, Nexenta has Inkscape, and I don't even see mention of Nexenta in the bug tracker for Inkscape.)

That's not to say that work doesn't remain to be done, but I think it does negate many of the reasons our editor suggests for sticking with Linux.

This effort deserves our respect

Posted May 17, 2007 8:31 UTC (Thu) by edschofield (guest, #39993) [Link] (6 responses)

I think this article is unfair. Choice is generally seen as a strength of free software, be it in filesystems, database servers, desktop environments, or distributions. Why should distributions with the Solaris kernel not be welcomed into the flock? If Ian Murdock and the OpenSolaris community can build upon the apparent success of efforts like Nexenta and produce a Solabuntu distribution with a Solaris kernel, Solaris userspace tools, and all our usual software that's "free, alive, fun, and clearly going places", the free software community will be the richer for it.

This effort deserves our respect

Posted May 17, 2007 12:08 UTC (Thu) by corbet (editor, #1) [Link] (5 responses)

Did I say anything about not welcoming them? Don't think so. What I did say is that this effort may not be enough to give Solaris the prominence that its developers wish it had. That's all.

This effort deserves our respect

Posted May 17, 2007 19:02 UTC (Thu) by edschofield (guest, #39993) [Link] (4 responses)

I might have misinterpreted the article, but it seems a little cynical. It seems to me that imitating the best aspects of current Linux distributions is exactly what Solaris needs in order to regain its past prominence. Solaris could still differentiate itself in other ways -- for example, through a focus on stability over novelty, as with the in-kernel interfaces, or through the weight of Sun's Solaris-related patents as a questionable source of comfort to the timid.

This effort deserves our respect

Posted May 17, 2007 22:42 UTC (Thu) by man_ls (guest, #15091) [Link] (3 responses)

It seems to me that imitating the best aspects of current Linux distributions is exactly what Solaris needs in order to regain its past prominence.

That is exactly what our editor says won't work :D Remember that Sun has already failed in that area, with the infamous Java Desktop System. Do you think that sticking a Solaris kernel inside will make much difference?

Solaris could still differentiate itself in other ways -- for example, through a focus on stability over novelty

Linux already has distributions focused on stability: Debian, Red Hat, CentOS.

as with the in-kernel interfaces

This one is better, but most people do not seem to be running around screaming for a stable ABI. Still, some do.

or through the weight of Sun's Solaris-related patents as a questionable source of comfort to the timid.

Worst option of all. This would make them lose the support of GNU/Linux communities, look at what has happened to Novell. By the way, Linux already has something like that (but community friendly) in the OIN.

If you ask me, Sun should focus on what they do well: solid engineering. Let them build a good kernel. Others may even do the packaging and distribution for them, with a little help; that is what the community does best. Then, when people go out looking for solid hardware they will hopefully not turn to IBM but to Sun.

In Kernel API

Posted May 18, 2007 17:26 UTC (Fri) by alex (subscriber, #1355) [Link] (2 responses)

Do you mean for people writing kernel modules or for the interface to user space?

The Userspace->Kernel API has never been fixed. To keep Solaris compatibility between versions you have to code to Sun's defined library interface.

In Kernel API

Posted May 18, 2007 18:42 UTC (Fri) by man_ls (guest, #15091) [Link] (1 responses)

To tell you the truth, I don't use Solaris so I'm not completely sure. Just from hearsay, I gather that edschofield refers to the kernel interface for writing modules, which is what Linux doesn't have and you hear certain people complaining about.

I didn't know that the userspace API was not fixed. So Sun's ABI compatibility claims refer to that library interface? Strange.

In Kernel API

Posted May 18, 2007 20:07 UTC (Fri) by alex (subscriber, #1355) [Link]

To be fair the syscall layer on both Linux and Solaris are pretty solid and stable. It would probably be fair to call Solaris's user space API as "System API" which is stable from release to release. You can run "system API" Solaris 8 binaries on Solaris 10 without any problems.

As far as in "in kernel" API for drivers and modules the strength of Linux come from it's ability to change and improve. The pain of tracking the driver API is considerably less if your driver is in the main kernel tree as it will get incrementally updated.

The sincerest form of flattery

Posted May 17, 2007 9:08 UTC (Thu) by rwmj (subscriber, #5474) [Link] (5 responses)

Before there was a Linux system in a reasonably usable state, the best system to have on one's desk usually came from Sun. As soon as it came in the door, however, it would be loaded up with crucial packages like the X Window System, gcc, netrek, emacs, and so on.

I'm not sure when the author is referring to. Linux really became usable around 1994. Before that if you really wanted "the best system to have on one's desk" you'd have chosen a NeXT cube or perhaps an SGI box.

After 1994 Sun systems just got worse in comparison to Linux systems. Around 1996 I benchmarked some Sun servers vs some FreeBSD and Linux servers. The Sun servers on a price/performance basis were 3 *times* worse than the PC-based servers. Sun desktops were largely based on PC components, so no more reliable. And they didn't have packaging systems which made installing and upgrading software an incredible chore. Not to mention the fact that they didn't come with any decent software - shells, compilers, etc. - in the first place. You had to get that from third party sites, and guess what it was just the GNU stuff which came with your average Linux distro in the first place.

Rich.

The sincerest form of flattery

Posted May 17, 2007 9:40 UTC (Thu) by tialaramex (subscriber, #21167) [Link] (1 responses)

The author is half-joking talking about the GNU userspace. What people would do (I didn't have the money, but soon afterward I met people who had at the time) was buy either Sun workstations, or cheap clones that came with Solaris / SunOS, and then they'd throw away all the horrible programs written in the 1970s and 1980s in K&R C and never properly maintained by Sun. To replace that they would install the GNU userspace, that means the toolchain (GNU CC etc.) and the text utilities, the editor (Emacs) etc.

You wouldn't do this to a NeXT box, because there were no (and despite effort, still aren't) better GNU replacements for most of the unique NeXT components. If you liked NeXT you were in for a long wait until Apple bought Steve Jobs and NeXT along with him.

You could do it to an SGI, but everyone I knew with an SGI workstation used it primarily for visualisation, with proprietary software. The SGI servers often had a lot of GNU tools installed though.

So the result is GNU/SunOS, and the arrival of good enough Linux kernels meant that you could replace GNU/SunOS with GNU/Linux and buy cheaper x86 desktop hardware. The experience for the user was very similar, and even for the administrator it was just a matter of learning a few new and differently arcane commands.

The sincerest form of flattery

Posted May 17, 2007 13:40 UTC (Thu) by nix (subscriber, #2304) [Link]

Some of us are still doing this. The Solaris development systems at my
workplace is running an elaborate, um, think of it as a per-process
packaging system whose sole purpose is to allow people to use the GNU
stuff they're used to *except* when testing things, whereupon they can
turn most of it off to get the savage ancient wreck that is Solaris 2.8
userspace, on which their programs will actually have to run in
production.

My impression is that my workplace's (very large financial) customers at
least chose Solaris for one reason: Oracle recommended it. Now they're
recommending Linux, *every* customer has jumped ship or is making plans to
do so, at least for the systems our application runs on. (Nobody is
planning to jump to Oracle's Linux distro.)

The Best System

Posted May 17, 2007 12:13 UTC (Thu) by corbet (editor, #1) [Link] (1 responses)

The "best" system was, of course, a matter of debate back then too. But I was not the only one who tended toward Sun systems. SGI boxes had nice hardware, but they were expensive and Irix was not as nice an experience. NeXT systems were'nt really practical in the environment where I worked. Obviously, in different situations, preferences were different, but the practice of replacing much of the software was the same.

As for when: in my work environment we started deploying Linux systems in production situations (carefully) around 1994.

The Best System

Posted May 17, 2007 13:42 UTC (Thu) by nix (subscriber, #2304) [Link]

In the more conservative big-City-of-London financial world I saw Linux
boxes start turning up outside experimental testbeds in 1999. By 2004 they
were everywhere.

Sun vs Linux

Posted May 17, 2007 22:24 UTC (Thu) by rfunk (subscriber, #4054) [Link]

I started running Linux personally in about 1995. In late 1996 I got a university job
supporting Sun workstations, where I worked until about 2000. As far as I could tell,
Solaris had just started getting good around 1995-1996 (whenever 2.6 was released).
And the machines were rock-solid at that point.

Jon's right -- everyone would boot up SunOS or Solaris, then add lots of free software,
either self-compiled or as a package from someplace like sunfreeware.com. (Yes, Solaris
did/does have a packaging system.) I even made a web page to keep track of it all; it's
still up, though extremely out of date: http://8help.osu.edu/wks/software
Eventually, probably after seeing sunfreeware.com, Sun started including a CD of
easily-installable free software packages with the OS.

Hmm, I wonder if your memory and experience (and benchmarks) are more from the
SunOS 4 era, while mine's more from the Solaris 2 era. (Or maybe you ran a really early
Solaris 2, when it was really bad.)

The Sun hardware didn't start getting PC-like until the Ultra 5 and Ultra 10 were released,
around 1998 or so I think. That was around the time my campus clients were starting to
move from Sun to Linux.

I also supported SGIs and had one on my desk, but they were more expensive, more
specialized, more proprietary, and generally harder to work with. And NeXTs were cool
but slow; one of the first web servers I messed with was a NeXT in 1996. Suns (and
Solaris) were general-purpose workhorses.

Somewhat off-topic, but I'm amused that I can now get a tiny Linksys NSLU2, put Debian
on it, and have about as much computing power as some of those workstations had a
decade ago.

The sincerest form of flattery

Posted May 17, 2007 9:55 UTC (Thu) by micka (subscriber, #38720) [Link] (2 responses)

> but it appears that the OpenSolaris community listens to Jörg about as much as the Linux
> community does.

Excellent!

The sincerest form of flattery

Posted May 17, 2007 13:44 UTC (Thu) by nix (subscriber, #2304) [Link] (1 responses)

Y'know, I feel rather sorry for Joerg. He so plainly doesn't know how to
communicate with other human beings except on an adversarial basis... he
must be a terribly lonely man. A good few of us can probably say that we
have few friends we didn't meet on the net, but I'd be surprised if Joerg
has even those.

I suspect his job is his life: hence the wild defensiveness when anyone
dares show competence in what he considers his personal field.

The sincerest form of flattery

Posted May 25, 2007 1:37 UTC (Fri) by wolfrider (guest, #3105) [Link]

--I feel for him as well; he may be an irascible German, but he does make some good points. cdrecord was *the* groundbreaking must-have for cross-system standardized CD burning not all that long ago.

Solaris is great

Posted May 17, 2007 10:14 UTC (Thu) by and (guest, #2883) [Link] (7 responses)

Whilst I probably won't give the Solaris kernel a closer look in the near
term future, I think it's good to have the prospect of choice when it
comes to the kernel. IMHO that is what really distinguishes the free
software community from the proprietary world. That's why I would be happy
if Solaris would bring something like the coopetition between Gnome and
KDE to the kernel in the future. As an added bonus, if Sun is really going
to use most of GNU's userspace in the future, this means more
contributions to that userspace which also helps people sticking with the
linux kernel.

Solaris is great

Posted May 17, 2007 13:14 UTC (Thu) by jospoortvliet (guest, #33164) [Link] (6 responses)

I agree. Competition is good. I think the competition between KDE and
Gnome is one of the main driving factors behind their growing success, as
they often share design ideas and sometimes even code. And they sure learn
from each other's experience, be it that they might be behind the other
one for a few years in some area's. EG KDE is now coming around on the
usability issue, and it's going to be a major focal point for KDE 4. On
the other hand, Gnome is slowly adopting KDE tech (mostly in a newer
incarnation) like DCOP (dbus) or changing their internal tech to
best-practices from KDE. Well, I actually can't come up with much examples
from the latter, but I'm not that much into Gnome - but I hope and think
they learn as well ;-)

Well, I actually HAVE seen a few discussions where there was cooperation
and mutual learning, so things aren't gloomy at all.

Solaris is great

Posted May 17, 2007 13:46 UTC (Thu) by nix (subscriber, #2304) [Link] (5 responses)

DCOP isn't KDE tech.

(And I certainly hope that KDE doesn't adopt usability stuff in the same
way GNOME has. I was forced to jump ship from GNOME to KDE entirely
because the `usability' improvements left me with a system I couldn't
customize to work the way I wanted it to, and I don't like to work inside
an electronic jail.)

Solaris is great

Posted May 17, 2007 16:28 UTC (Thu) by jhs (guest, #12429) [Link] (2 responses)

The best joke I've heard about GNOME usability is that eventually its interface will evolve to be so simple and "user-friendly" that there will be no more buttons or input: we will all just sit and stare at the screen, while GNOME presents us with things it knows we want to see and do.

(Actually I am a GNOME user, mostly because it's the default in most distros, and as long as I have the few basic context-switching keyboard shortcuts I'm good to go. On that note, I wish freedesktop.org or somebody would put effort into unifying the many types of context-switching activities: virtual desktops, firefox tabs, terminal emulator tabs, GNU screen windows -- I think it's time to draw up some reasonable abstraction about this and simplify it.)

Solaris is great

Posted May 17, 2007 22:25 UTC (Thu) by nix (subscriber, #2304) [Link] (1 responses)

Unifying those would be tricky, but providing an interface that allows
external programs to query and (to a limited extent) command them might be
doable.

(I mean, how could you unify a Firefox tab --- restricted to displaying
Gecko-rendered content in X --- with a GNU screen, uh, screen, which is
textual and can be displayed on a dozen displays at once and has an
associated pty, program, even an associated logged-in user? They're
completely different things. They don't even share significant amounts of
related queryable state. In fact I can't think of *any* state they share.)

Solaris is great

Posted May 23, 2007 22:25 UTC (Wed) by njs (subscriber, #40338) [Link]

They're all rectangular areas that the user wants to look at, resize, switch between, etc. Firefox and terminal emulator tabs are really just hacks around the fact that the standard "desktop metaphor" window managers suck at actually organizing information in a useful way. If you use a tabbing wm, then suddenly all of your apps are effectively tabbing, with a single UI... screen is a bit of a special case, but there's no reason one couldn't have a screen client that blew up the multiple ptys into multiple X windows.

What you lose right now if you do this is all the little details -- Firefox tabs show favicons and page load throbbers (or in some versions, even show progress bars and color code the title text), screen will tell you when a bell goes off in another terminal, etc. (Also, all existing tabbing wms are also written by minimalist freaks against Xlib directly, and thus not very pretty or featureful. I've been known to be a minimalist freak myself, but it's not going to get you app developer buy-in these days, and minimalism and tabbing are totally orthogonal dimensions. I've been seriously tempted to write a full-on GTK-or-Qt+XRender+XComposite-using tabbing wm myself, anyone want to help?)

Where an API would be useful is in exporting this information in a generic way, so that wm-provided tabs could give you just as much contextual cues and status information as program-internal tabs. There's actually a vestigial hint of such a system in the "urgency" hint that ICCCM defines, that makes your IM windows throb when someone says something to you -- that's enough maybe for screen's bell notifications, but we could go much further.

Solaris is great

Posted May 24, 2007 10:11 UTC (Thu) by quintesse (guest, #14569) [Link] (1 responses)

"DCOP isn't KDE tech"

Are you sure? Reading this message on the KDE website from 1999 it seems pretty much something that was made especially for KDE: http://developer.kde.org/documentation/other/dcop.html

Solaris is great

Posted May 24, 2007 19:17 UTC (Thu) by nix (subscriber, #2304) [Link]

Sorry, mental hash collision between DCOP and DBUS. DCOP is indeed KDE
stuff (the source is in kdelibs).

Debian!

Posted May 17, 2007 10:29 UTC (Thu) by smurf (subscriber, #17840) [Link] (5 responses)

Well, the inventor of Debian should have no problem whatsoever porting the whole of Debian to OpenSolaris. Problem solved.

It's been done before (with FreeBSD and the Hurd), so ...

Debian!

Posted May 17, 2007 11:54 UTC (Thu) by jengelh (subscriber, #33263) [Link]

...eventually it all ends up in the melting pot.

Debian!

Posted May 17, 2007 14:01 UTC (Thu) by jeroen (guest, #12372) [Link] (3 responses)

And that work has been done by actually not that a big number of volunteers. If Sun would simply hire 10 Debian developers and let them work on porting Debian to Solaris full-time they would probably have a working Debian GNU/Solaris system with the release of lenny.

Wouldn't it be nice if the Debian installer would ask which kernel you want to install? It would really be the "universal operating system".

opensolaris-image.deb

Posted May 18, 2007 7:58 UTC (Fri) by xoddam (subscriber, #2322) [Link]

> Wouldn't it be nice if the Debian installer would ask which kernel
> you want to install?

Already in the last release cycle they changed the name of
the "kernel-image" packages to "linux-image"; when setting up a new system
with debootstrap the last manual step is to install a kernel and
bootloader. I'm *certain* that opensolaris-image.deb could be packaged
analogously for the consumption of grub, though I'm sure there are
userspace packages which are insufficiently kernel-agnostic to allow such
a simple switch.

The nexenta GNU/solaris port (http://www.gnusolaris.org/) doesn't make
things quite this esay, but the principle is already established.

Debian!

Posted May 19, 2007 5:33 UTC (Sat) by snitm (guest, #4031) [Link]

I will not be surprised if Ian Murdock hires Progeny refuges to work for Sun to do just that...

Debian!

Posted Jun 4, 2007 9:53 UTC (Mon) by csamuel (✭ supporter ✭, #2624) [Link]

Erm, would that work ? My understanding is that there is a question
whether the CDDL would pass the Debian Free Software Guidelines
(especially the choice of venue and non-anonymity rules in the CDDL).

If it didn't pass the tests then the OpenSolaris kernel would end up in
non-free, and the Debian Social Contract says Debian will never depend
upon non-free software...

Device drivers and interfaces

Posted May 17, 2007 12:18 UTC (Thu) by copsewood (subscriber, #199) [Link] (5 responses)

Availability and quality of hardware drivers for the 2 kernels is likely to be a significant factor determining the usefulness of these 2 possible kernels to most potential users. It will be interesting to see how quickly if ever standardised device interfaces allow a driver developed for kernel A to work well on kernel B. This might work already for some classes of device, e.g. many USB storage devices seem to use common enough protocols for you to be able to plug these in on Linux and they just work. Compatibility layers and taking device drivers into user space seems likely to impact performance but might enable more devices to be supported. For servers which are less likely to need to support a great and changing variety of peripherals, raw kernel performance is more likely to factor in software choices.

Device drivers and interfaces

Posted May 17, 2007 12:23 UTC (Thu) by mjg59 (subscriber, #23239) [Link] (3 responses)

Indeed. While there's the argument that the CDDL makes life difficult for Linux (we can't just steal the existing zfs implementation), it's also an issue for Sun - there's no way to make use of the vast number of existing Linux drivers, and so they'll have to rewrite many of them from scratch. That's a huge duplication of effort, and no matter how much of a bonus dtrace and zfs might be there's little incentive to use opensolaris if it doesn't drive most of your hardware.

Device drivers and interfaces

Posted May 17, 2007 13:20 UTC (Thu) by ewan (guest, #5533) [Link] (2 responses)

For now. If Sun move OpenSolaris to GPLv3 when it's ready then they'll be
able to lift any 'v2 or later' code from Linux (of which there's quite a
bit), without letting anything flow back to Linux from OpenSolaris.

Device drivers and interfaces

Posted May 18, 2007 4:16 UTC (Fri) by dlang (guest, #313) [Link] (1 responses)

and if they started doing that I'll bet far more device driver authors would change their future code to GPLv2 only.

letting someone take your code, change it, but then not being able to use the changes in the original is exactly the type of behavior the GPL is supposed to prevent,

Device drivers and interfaces

Posted May 24, 2007 12:00 UTC (Thu) by zotz (guest, #26117) [Link]

" and if they started doing that I'll bet far more device driver authors would change their future code to GPLv2 only."

I am not so sure about that. If they are OK with the GPLv3, they may instead bring more pressure on the remaining GPLv2 people.

all the best,

drew

Device drivers and interfaces

Posted May 18, 2007 15:29 UTC (Fri) by cajal (guest, #4167) [Link]

"It will be interesting to see how quickly if ever standardised device interfaces allow a driver developed for kernel A to work well on kernel B."

If that were to ever happen for Linux, it would require the Linux kernel to adopt a stable, in-kernel driver interface. The kernel developers don't seem to want that - http://www.kroah.com/log/linux/stable_api_nonsense.html.

Such driver cross-adoption does happen frequently between the BSDs, though, since their driver APIs are fairly similar.

6 months release cycle

Posted May 17, 2007 20:08 UTC (Thu) by mikov (guest, #33179) [Link] (4 responses)

(Off-topic warning)

Is 6 months release cycle really such a good idea ? To me 6 months doesn't seem enough for serious testing of a distribution.

This may not mean anything, but during the recent upgrade to Kubuntu Feisty, I hit three different upgrade-related bugs while upgrading two machines. All of the bugs I would qualify as serious. Two different (and unrelated) crashes in the upgrade tool - there were reports of unusable systems after that, but I wasn't that unlucky myself. The third bug was destroyed Konqueror profiles making Konqueror unusable (missing menus, etc).

I was disappointed but it wasn't a big deal for me - this is what makes Linux fun to a level. Unfortunately I was seriously embarrassed because one of the upgrades was on a friend's machine, who I just recently persuaded to move to Linux. He told me directly that he did not think going through this (manually patching upgrade scripts, struggling to recover Konqueror) every six months was acceptable. I agree.

I can't imagine how thousands of non-hackers would deal with this if they started buying from Linux machines from Dell and had the same problems.

Instead of making absurdly short release cycles, distributions should release new supported versions of packages for the current distribution. There is no reason whatsoever that one should have to upgrade his OS and all other packages, only so he would get the newer version of one package that he cares about.

6 months release cycle

Posted May 18, 2007 11:50 UTC (Fri) by joyride (guest, #4606) [Link] (3 responses)

I was just about to post a comment on this to (and I now do :). The reason I have started going back to solaris is that they dont have 6 months release cycles. Solaris have 10 year support for every release I can install solaris 10 and still run it 2010 and have no problems. I do have solaris 2.51 boxes still, unfortunately those are too old for support.

I don't want to reinstall a 100 servers every 6 month. Maybe every 3rd or 4th year at most.

6 months release cycle

Posted May 18, 2007 16:07 UTC (Fri) by mikov (guest, #33179) [Link] (1 responses)

I feel your pain. 6 month upgrade cycle for Solaris seems absolutely silly.

Actually, for servers the Ubuntu situation is not that bad, because they have the LTS release, which is good for at least a couple of years. And I'd use Debian for servers anyway.

But! This IMHO _broken_ trend towards shorter releases seems to be very popular and even Debian is apparently trying to switch to it. If I have to upgrade Debian servers every year, that would be terrible. Not only that - we use Debian for embedded projects, where upgrades are out of the question, but it is still preferable not to stay too far behind the current distribution.

Also, the truth is, contrary to what most people assume, frequent releases are even bigger problem on the desktop than the server, because:
- There are more desktops that servers
- Desktops are not used by professional sys-admins or hackers, so even the smallest glitch is a huge problem

With Windows there is no danger of too frequent releases :-)

6 months release cycle

Posted May 22, 2007 18:28 UTC (Tue) by k8to (guest, #15413) [Link]

At least with Debian, their policy is to provide security fixes for stable releases X months after the release of the next stable. (I think it's 12 months?)

Debian would like to aim towards shorter release cycles... like every year, instead of every 18-24 months. So you would have easily installed fixes and security patches for around 2 years at least.

Granted this isn't terribly long compared to some unixes, but it's not horrible, especially when you consider that it is plausible to run servers entirely out of packages, and that upgrades are semi-automated.

But for some servers I would definitely prefer systems with longer support guarantees. Heck, I might even pay for them :-)

6 months release cycle

Posted May 18, 2007 22:40 UTC (Fri) by raven667 (subscriber, #5198) [Link]

I don't want to reinstall a 100 servers every 6 month. Maybe every 3rd or 4th year at most.

I agree and this is one of the reasons many of the systems I manage are running RHEL3, heck some are still running RH7x). I think there is a significant difference here though, Ian comes from the Debian camp which has a great history of supporting online in-place upgrades between versions. When I set up a server that is not going to have regular on-site maintenance I prefer Debian for this reason, I can keep it current for years including major upgrades without worrying that I'm going to brick a host that's inconvenient to get to and fix. I haven't had that experience with other distros and don't generally trust their upgrade systems.

The people I know who use Ubuntu were pretty impressed by this last upgrade which from what I can tell involved clicking an "Upgrade" button in the software manager and waiting until the progress bar hit 100% then rebooting into their new desktop. It couldn't be any easier.

So really a 6mo release cycle isn't a problem if the update system works well and the distributor avoids making radical changes that break everything but instead treat each new version like a service pack or refresh of the same system. Allowing one to keep up with current software without requiring major disruption every couple of years seems ideal to me.