Slashdot Mirror
Outlining Thin Linux
snydeq writes: Deep End's Paul Venezia follows up his call for splitting Linux distros in two by arguing that the new shape of the Linux server is thin, light, and fine-tuned to a single purpose. "Those of us who build and maintain large-scale Linux infrastructures would be happy to see a highly specific, highly stable mainstream distro that had no desktop package or dependency support whatsoever, so was not beholden to architectural changes made due to desktop package requirements. When you're rolling out a few hundred Linux VMs locally, in the cloud, or both, you won't manually log into them, much less need any type of graphical support. Frankly, you could lose the framebuffer too; it wouldn't matter unless you were running certain tests," Venezia writes. "It's only a matter of time before a Linux distribution that caters solely to these considerations becomes mainstream and is offered alongside more traditional distributions."
I'm sure I've installed minimal gentoo and Debian systems that fit that description.
Nullius in verba
I'm sure I'm also wrong somehow. I haven't touched Slack in 10 years. What am I missing?
I think this makes a lot of sense. I don't know if I would call it "Splitting Linux in two", but just offering server installations separate from desktop installations (Subuntu? Linux Smint?) would do it.
Same OS, different uses, different default programs. Sounds like the unified Linux way to me.
http://www.linuxfromscratch.or... Everything you need, nothing you want.
Pretty much fits the desc.. Supports docker & clustering but very very little else. Should probably shoe-horn a hypervisor onto that little OS.
I see this as a response to the systemd war, and a viable one at that. A server does not need systemd... "It boots faster." Why bother when post takes 20 minutes? "It is tied into udev and network manager." Servers generally don't dhcp or hotplug... Since "the desktop" is going full tilt boogie in one direction and damn everyone who disagrees, it makes sense for the server folks to say "See ya!" And soon after someone posts about how to get lxde running on the server. :)
Windows Server 2012 R2 Core? What about Ubuntu JeOS? (Just Enough OS)
Ken
If you don't like something don't put it in the container!
DeepEnd's Paul Venezia needs to hire someone who knows what he is doing instead of telling people what they should do for his amusement. Fedora and co:
yum -y install @Core
You don't even get tar. Only recently did you get scp.
All the major Linux distributions already do this. Ubuntu has a minimal and a Server spin, Fedora has a new Server only spin, CentOS and RHEL have Server only spins, Debian has a minimal install, etc etc etc. The only way the guy's arguement makes even a little sense is if he thinks a server distribution is somehow made better by it not being possible to add a desktop interface to it. In other words, a distribution that ships only server packages and refuses to ever include anything that features a GUI.
Even then the arguement doesn't make any sense because it would assume a distribution loses polish in one area if it also allows packages in another area. There is no reason to believe this is true.
Takes some effort to configure, but small, fast, and light on resources.
http://tinycorelinux.net/downloads.html
My proceedure for weekly security updates for my computer is: download all packages to be upgraded, then reboot into maintenance mode and physically install them. Basically I run LMDE (debian unstable mostly).
I occasionally check the downloads and see a strange package being downloaded. So I look up the dependencies, and see some of the most bizarre dependencies. For example, that KDE requires CUPS ( I don't have a usable printer ).
So instead of worrying about what needs to be installed first clean up all the dependencies so you don't install packages you don't need or want.
I myself once proposed to have the desktop separated from the server by suggesting it to a guy who was going to make his own Linux distribution.
Alas, this happens naturally as many distros focus on the server side.
But Paul has a valid point, which is what I've been thinking, too: the needs of a server are way too different from those of a desktop. And Linus & Co., while they recognize the desktop problem and even complain about it, are not desktop guys. Which is nice, they pretty much made everything work fine at the kernel and let supercomputing, embedded etc. for specialized dudes.
My current thinking is somewhat mixed, though: I mostly use the desktop (and it has become really nice recently), but I have and plan to add servers for home uses. I wish someone put more desktop inside servers, making them more intuitive, with more hand-holding -- the kind of server a small architecture office would want.
Since I have a personal history in IT, I managed to configure samba, NFS, squid, cups and a firewall... but it's not easy for a non-expert like me. Server UIs could provide a basic version a newbie could use. Heck, just configuring the network is hell for the likes of me. I could use a graphical tool, maybe the same to configure the firewall...
Reading this it feels like removing tab completion in the shell is something to sought after. It's for interactive use and needlessly weighs down thin and lean VM server installs.
Make sure support for ncurses and ssh -X is eradicated, too.
"Thinux" surely?
Linux sucks. Windows isn't much better other than the support.
Computing is not where I thought it would be 25 years ago. Users have continually less power, not more.
Linux gives people power in the wrong places. Places people rather let the system do the work. And it's based on Unix and Unix frankly sucks.
Fuck. I wish Plan 9 or Lisp Machines or something else won other than this half-ass kludge.
Fuck it. I'm going to sell my house tomorrow and build a log cabin in Canadian woods before the winter arrives. Out of here, bitches.
I thought LFS was pretty much such a system.
Just read Lennart Poettering's article the other day on convergence/brtfs. After having overcome some reflexive
repugnancy, saw that it made sense, and O got interested to see how that will work out.
Now Paul Venezia makes the case for divergence, in a particular (though not unimportant) use case. Makes sense as well!
I guess all this versatility is inherent in Linux -- it's just the issue at which point the di- or con- is applied, and who does
the applying, so to speak.
Not trolling... I don't use BSD really, but my understanding is that some of the BSD distros are more server focused. I don't mind being corrected but my understanding is this could be a legit alternative if the idea of splitting Linux is a no go. I don't know why BSD isn't seen or heard of more (I do know it is used and has a strong following, but doesn't seem as prevalent as Linux... Mac doesn't count here). For BSD adherents, maybe this is the break they are looking for?
-- I ignore anonymous replies to my comments and postings.
Every server is different, packages and dependencies are very much relevant. Some need a fully functional framebuffer with OpenGL support to generate web images/video.
If you are asking to configure system on build master and then mass deploy to individual servers, without any unnecessary development/configuration tools, that's a reasonable idea.
We used to run linux in the server room because it was lean and easy to admin. Windows was slow, mousy, and dependencies were hellish.
Now we run Windows Server 2012 with no GUI, virtualized, and admin with powershell. We've ripped out tens of thousands of dollars of Red Hat; windows is cheaper.
Basically there aren't any linux server distros that are like Red Hat used to be before the Fedora fiasco. It seems like Red Hat today is doing a bad job of trying to be a GUI laptop distro running on server hardware. And they are letting mature stuff like PADL's LDAP modules go to seed while shipping raw, buggy stuff like SSSD, instead of maintaining the old stuff until the new is reliable enough for real world use.
Single purpose machines need a single purpose OS and kernel. All the scheduling and protections offered by linux are currently offered by your hypervisor. A new OS like mirage (http://www.xenproject.org/developers/teams/mirage-os.html) or osv (http://osv.io/)
Yes I would like this. Thank you.
We can call it Thinux.
He points out, correctly, that many servers don't need much. Particularly with cloud services, servers might spin up a whole bunch of very lightweight virtual machines doing one thing (running a web server like nginx for example).
So his big idea is a "server-only" distribution that doesn't have any support at all for GUI operation. But he doesn't really explain the benefit. As far as I can understand, he names one single benefit: such a distro would be "not beholden to architectural changes made due to desktop package requirements."
The only "architectural changes" I can think of recently are related to systemd, so I guess this was his very roundabout way of wishing for a Linux distribution with no systemd support.
Am I wrong here? Did you manage to find any other advantage listed in his article to explain why it would be great if you were unable to set up a machine running your server OS with a GUI?
lf(1): it's like ls(1) but sorts filenames by extension, tersely
Not just thin. The time has come for cluster-targetted distributions. Openstack, CoreOS and others are minimal linux meant to do mostly a supporting work for loading over them as VMs or container clouds the more bulky application linux servers/images/containers. All is about having a bunch of servers (real or virtual), installing something minimal that builds a cloud on them. It's linux all way down.
One thing we need to do is pry open the BIOS. Why should we need a VGA and keyboard or serial port to configure the hardware? Why is initializing the memory interface a secret? We can't improve servers if the hardware is still tied to the original IBM PC.
Basically he wants Linux distro to morph into NetBSD (or whatever other BSD fits you better, NetBSD is probably the most obscure but has a few very interesting points: excellent backward compatibility, cross-compilable out of the box)
Admins of systems that serve serious stuff will know how to secure their systems, but the box-or-two business will be running back to Windows (with good reason) due to the horror of not wearing a bloody useless cycle helmet - oh I mean not logging ion as root.
Home-linux distros needs nil access restrictions then good advice. Security depends on people so EMPOWER PEOPLE by EDUCATING them why layers of security are good and worth worrying about. SUDO is box-ticking for Linux.
You may be right, but I wonder if the author of the article is aware that one of the leading cloud friendly distros, CoreOS, uses systemd. If fact, systemd is an integral part of fleet:
RedHat's geard, which is part of OpenShift, also uses systemd.
It seems to me that the opposite is happening, cloud ready distros are choosing systemd.
It's only a matter of time until lazy-assed system admins create installation packages (aka "distros") that include exactly what should be running on a particular VM.
I mean, FFS, what the hell does the OP expect? For the world to do his job for him?
Lazy bastard...
I do not fail; I succeed at finding out what does not work.
PCLinuxOS
It's his job to find some random controversy that gets traffic. Note that this series of articles is all coming from one person and slowly ratcheting up as he finds out how reactive some readers are.
"There are some people that if they don't know, you can't tell them." ~ Louis Armstrong
I am sure some people are interested in the topic on its own right, but obviously most just want to escape from systemd. In this case, please first read this paper. Even if you think the end result is crap, there are some very lucid ideas in there. You would do well to at least consider them as you adopt OpenRC or whatever in your thin Linux distribution. I swear that I have no relationship with systemd project and only occasional hobbyist relationship with Linux.
Embedded distros are extremely thin and server distros don't have desktop components, so I don't know what this guy is complaining about. It looks like he has never heard of Anaconda and Kickstart.
Excuse me, but please get off my Pennisetum Clandestinum, eh!
Throw out the vga hardware, usb, keyboards, mice, etc. That stuff really isn't needed on servers. Clean up a lot of other 'personal computer' stuff and grow up.
Throw out systemd and Gnome and NetworkManager and SELinux. They break things, and the return on investment for the disk and RAM the use is non-existent.
FreeBSD is what I'd pick.
www.openbsd.org
Because MYAPP can run on any of my nodes, except for nodes running HISAPP, or HERAPP, but plays nicely with THISAPP.
Fleetctl and systemd is part of the recipe.
We just need to make building your own distro much faster and easier. Rather than installing packages, a tool should automatically rebuild and redeploy.
I used to use Gentoo minimal installs but recently discovered Alpine Linux (http://alpinelinux.org/) which is even better.
With respect, the above poster is replying to someone that appears to be asserting that. I suggest reading other posts higher up in the thread before wasting time writing such long replies that miss the point.
This is a silly article.
What this guy is looking for is called BSD. In the past, base system was bit less than 30MB. Useless, but still less than base setup of most modern distros.
Among Linuxes, probably only Slackware stayed relatively close to the roots and still can be stripped to the bone. And Debian isn't that far off, really, if you are willing to go on rampage with the rm command (remove man pages, documentation, supplemental files, localizations, etc).
Othereise, this guy has probably missed completely that people are already for years building their own "lean and slim" special-purpose distros using the Gentoo as a factory distro. Because what he asks is really "special-purpose". In most real-world cases, the disk space is cheap and the users want to be able to install new software with just few clicks.
All hope abandon ye who enter here.
Eh?
http://www.youtube.com/watch?v=sZa26_esLBE
It sounds like he's talking about a pain-in-the-ass problem he's having to deal with, so I'm not going to flame him for that. If this is what works for him, that's great.
But what about the other 99% of Linux installers/administrators? You know, the rest of us, who are rolling out non-VMs and doing them one at a time?
I can fix this by changing "the new shape of the Linux server is.." to "a new shape of the Linux server is.." See how easy that was? You just need your special distribution, for your special problem.
None of this is evidence that Linux distributions need to be "split" into two kinds; I think it's more like it's evidence that Linux's 43 distributions might need to expand to 44-48.
Venezia, you're just another use case. I'm not putting your case down, but don't think of it as a big major thing, or as though suddenly most everyone is doing what you're doing. You're sounding a little like someone who sees cars as "We need to split the auto industry in two, because there's the Toyota Camry, and there's everything else." No, your Camry, as fine a car as it may be, isn't that important, and neither is your particular approach to servers.
This submission was made by snydeq who may or may not be Paul Venezia, but certainly appears to have a clear vested interest in frequently promoting Paul Venezia's column and other articles from Info World on a nearly weekly basis.
Considering the overwhelmingly poor quality of the vast majority of Info World's trade rag (slang trade magazine), where most of the better "articles" (i.e. aka "filler," the stuff between the ads) tend to be cribbed from vendor's white papers, don't seem to merit being frequently promoted at Slashdot unless there is a financial arrangement in place, in which case the ethics of journalism would indicate that such a financial arrangement should be disclosed to readers.
Not that I'm suggesting Slashdot considers itself involved in journalism, regardless of the usage of the terms such as: articles, submissions, and editors in the Slashdot vernacular. I will mention that the US FTC publishes March 2013 disclosure guidelines for sponsorship, marketing, and promotions.
Read it again. Praise for something is praise for something.
Nice little schoolboy touch with trying to turn things back onto me. If you wanted to make people think less of your worth with each post you are doing a fine job. You've convinced me that you are not just someone that has missed a point but instead someone that wants to actively spread misinformation.
I am fighting with Ubuntu and debian and the hell of it is that I like recent stuff, like Ipython 2 and Emacs 24, and yet debian and Ubuntu are awful bloatware. Most of the debian packages, at least the ones I've seen through Ubuntu are crap, half written non-orthogonal code. And I am just amazed what I can do inside emacs. I am beginning to think that I ought to revert to the command line and make emacs my GUI and forget all the other crap, including the browsers, which are all huge. And what really gets me upset is the cache files that gets put on my home dir by Google and Mozilla and countless other applications that come off the debian tree. Some idiot came out with a tiling GUI file display manager, and I kept thinking that has all been done in emacs 30 years ago; some idiot trying to reinvent something old.