[DE] Lennart Poetterings Vision zukünftiger Linux Distributionen

[devil]

Deutsch: http://www.pro-linux.de/news/1/21461/poetterings-vision-von-der-linux-distribution-der-zukunft.html
Original: http://0pointer.net/blog/revisiting-how-we-put-together-linux-systems.html


greetz
devil

[piper]

Great read danke

[michaa7]

I read it on Pro-Linux.

Although I am not using systemd I agree with the efforts to consolidate/simplificate Linux distribution developpment. Sounds very interesting.

[ralul]

Isn't it a statement to give up on consolidation but to recreate docker more efficiently ?

[devil]

Well, he picks the best of docker, coreOS and others to make packaging and distributing software genericaly easier. If they are able to follow their plan, that would make distributions move closer together without giving up their individuality.


greetz
devil

[ralul]

Well, he picks the best of docker, coreOS and others to make packaging and distributing software genericaly easier. If they are able to follow their plan, that would make distributions move closer together without giving up their individuality.

I understand this LP statement differently:
They want to make the packaging of software apps easier and distributing these apps of course in a perfect way. But this may result in diverging distributions, because there is no need to consolidate any more.

... the need to have generelized distributions, which can run all apps, may vanish, so distros can focus on their strengths. The only need is to have a common subset of basics and a common linux kernel. This will very much change the Linux ecosystem on the long run.

Thus: the linux kernel with a common subset of basics delivered by systemd would be a hypervisor to boot up many virtual machines: an Ubuntu unity as user interface, a debian machine to run scientific apps, an oracle machine to deliver the database server. Whatever a developer chooses a distribution to target his application.

[devil]

I agree other that using virtual machines. It would be containers, that are already revolutionizing packaging and deployment of apps as of today. Hence my hints to Docker and CoreOS. The latter already uses the stateless approach up to a point, when it has 2 root filesystems (only one readable at any given point in time) on one machine and runs automated updates even on big clusters. Should an update go wrong, you just switch the other filesystem to rw. I wrote something on this not too long ago for ComputerBase .


greetz
devil

[michaa7]

...
They want to make the packaging of software apps easier and distributing these apps of course in a perfect way. But this may result in diverging distributions, because there is no need to consolidate any more.
...

I didn't read it this way. To me it is absolutly not clear what the effects of this development will be in the long run. But I imagine it leeds or could lead to more modularized systems. They won't differ so much in individual libraries, but in sets of libraries which *could* have an consolidating effect. In this scenario distributions are confronted with developers how bind their application to a *defined* environement. It is then up to the maintainers whether or not they *unbundle* it. That's how I understand the situation. I think this can lead to much unification because now (then) maintainers need good reasons to give up cross distro solutions.

What makes me shaking my head is something else: I know of deduplication as far as storage is concerned. But in RAM you may need to load different similar environments in parallel. I wonder how much RAM we need in the future when in the *extrem case* potentially each application may come with it's own environment? Ok, "each" is really, really extrem, it would be insane, but possible, so let's say many applications.

[ralul]

@devil,
arrgh ... I got confused: Yes  virtual machines do have their own kernel. We speak about containers!
Up to now containers mostly have been of one distribution. Now LP speaks about a multi-distro container environment ...

But this may result in diverging distributions, because there is no need to consolidate any more....

I didn't read it this way. To me it is absolutly not clear what the effects of this development will be in the long run.

  Yes, we are speculating !

But I imagine it leeds or could lead to more modularized systems. They won't differ so much in individual libraries, but in sets of libraries which *could* have an consolidating effect. In this scenario distributions are confronted with developers how bind their application to a *defined* environement. It is then up to the maintainers whether or not they *unbundle* it. That's how I understand the situation. I think this can lead to much unification because now (then) maintainers need good reasons to give up cross distro solutions.

Yes, good reasons for meaningful changes!  That is where diversification should head to. Special purpose implementations!

But don't ask me what these will be in the future. I cannot imagine. In the past you would have implemented special Linux systems by using special Linux kernels with special filesystems. But the LP  vision needs to have one common kernel and one filesystem only!

My speculation: This LP vision makes it easy to lazily develop any application for just one distribution and not to take care any more  it will run on other distros also. Be prepared of upstream developers rejecting siduction patches:  The lazy and meaningless multitude will grow