More Debian GNU/Linux

Debian GNU/Linux 8, Jessie has been released. I made a download of the official “netinst” CD-image and used it with kvm to install a virtual machine (dd if=/dev/zero bs=1024k count=4096 > jessie.img;kvm -m 512 -boot order=d -cdrom debian-8.0.0-amd64-netinst.iso jessie.img). The installation took 10 minutes for a minimal installation and was uneventful. My first boot was a shock, however. I could not even log in thanks to systemd being unhappy (failed to start logind). I had to reboot to “rescue” mode to get in. When I first installed a GUI and started it with “service lightdm start”, I could not login again… Another reboot made it work.

Debian, please save us from systemd. If we wanted PCs that behaved like that other OS, we could just use TOOS… systemd, with a tiny GUI and TuxCommander running, uses 5.6MB RAM with 28MB virtual memory. 77 tasks running… Sigh…

OTOH, it did boot and shutdown like a rocket and XFCE4 was smooth and beautiful.

Source: Debian — Debian “jessie” Release Information

UPDATE I found Debian Bug report logs – #769728 Logind does not start properly on freshly installed minimal systems had been dealt with last winter. Perhaps this is a regression, logind not starting… Clearly, it’s not mainstream as most take some of the tasksel options. I have preferred to take none so that I have total control of the installation. Perhaps they need to redefine “minimal” by taking the basic utilies off the list. Few would notice with all this bloatware.

About Robert Pogson

I am a retired teacher in Canada. I taught in the subject areas where I have worked for almost forty years: maths, physics, chemistry and computers. I love hunting, fishing, picking berries and mushrooms, too.
This entry was posted in technology. Bookmark the permalink.

35 Responses to More Debian GNU/Linux

  1. ram says:

    Moore’s law or not, the trend now is to parallel computing. Yes, people were predicting this in the early 1980’s, about 30 odd years before it actually happened, but it did have to happen eventually.

  2. oiaohm says:

    That’s crap. I have two really slow PCs (Atoms) used as home theatre PCs on big screens with no problems at all on audio/video. They run Debian GNU/Linux. My Beast does nothing fancy with audio/video and it’s just fine. There’s nothing “real time” about any of those systems.
    Robert Pogson its not crap. Most people don’t notice the minor added audio glitch or the odd teared video frame. The new Nvidia and AMD drivers are supporting tear free on Linux now that the framework in kernel exists to support it. Audio tears are can be very much more noticeable than video ones.

    I said doing good Quality. Doing decent quality a glitch every 3 to 4 hours is like fine.

    Please note even dmix in Alsa itself can create a audio malfunction under the right conditions due to lack good IPC framework.

    luvr Ubuntu cross over mix is the same as Debian and it does open pandoras box at times using the service command instead of systemctl.

  3. DrLoser wrote, ““Moore’s Law,” Robert. It’s even more inappropriate in this case.”

    Oh, shut up. Moore’s law is very relevant. If Moore’s Law doubles the capability of the maker to put bits on a memory chip, it’s very likely that future PCs will have more memory to absorb whatever bloat/load systemd can dish up. Systemd has obviously grasped many parts of a GNU/Linux system but it’s not a growth industry so PCs will catch up or stay ahead of it. Even my ancient crates can afford a few percent of CPU time/MB of RAM to systemd. A modern system won’t even notice.

  4. DrLoser says:

    I’m pretty sure Moore’s Law, the GNU/Linux ecosystem etc. will accommodatedeal with systemd by then.

    I wish you’d stop parading your ignorance of “Moore’s Law,” Robert. It’s even more inappropriate in this case.

    At a sufficiently abstract level for this sort of discussion, systemd is nothing but an engine running on a declarative syntax to create a dependency chain.

    You don’t run the elements of a dependency chain in parallel. Moore’s Law isn’t going to make an appreciable difference to the systemd equivalent of your SysVinit setup over the next five years.

    The only thing stopping systemd working for you right now is your total inability to understand any computing concept that has emerged into the public eye since about 1980.

  5. I wrote, “That might be possible in the next two years.”

    I hadn’t realized that Jessie is a LTS release and I may be able to escape systemd on Beast for five years if I live that long… I’m pretty sure Moore’s Law, the GNU/Linux ecosystem etc. will accommodatedeal with systemd by then.

  6. ssorbom wrote, “I rather wonder how you upgraded”.

    I used apt-get dist-upgrade long before systemd was made the default init. Everything changed after that. A dist-upgrade might still work if one installs wheezy and then dist-upgrades to jessie. On a minimal installation, that’s not a huge burden. It’s just a few hundred packages to be installed twice.

  7. oiaohm wrote, “The IPC is one of the biggest reasons why Linux has not been readly for the desktop. Poor grade IPC there is no way todo good quality audio or graphics without droping to using real-time modes that are highly power consuming.”

    That’s crap. I have two really slow PCs (Atoms) used as home theatre PCs on big screens with no problems at all on audio/video. They run Debian GNU/Linux. My Beast does nothing fancy with audio/video and it’s just fine. There’s nothing “real time” about any of those systems.

  8. luvr says:

    Distrowatch has a feature story on the newly released Ubuntu 15.04, which appears to have implemented systemd in a satisfactory manner. A citation:

    Ubuntu has taken an approach I like with regards to adopting systemd. While systemd functions as init, the old methods of managing the system still exist on Ubuntu. For example, services can be started or stopped using either the old style “service” commands or systemd’s “systemctl” command. Ubuntu still maintains log files in text format which means daemons that monitor logs (such as security applications) do not need to be patched to work on Ubuntu. The systemd binary journal is also present, so people who want to take advantage of the journal’s search features can do that too.

  9. oiaohm says:

    But why isn’t dbus included in the debian-base meta-package?
    ssorbom dbus is in the default debian meta package but Robert by passed it. Linux systems are highly customisable so you can go into high risk areas if you choose to. There is a lot of stuff in the default meta package. To be truthful you can cut the default meta package in half and still get a fully working system. A lot is in the default meta package so closed source installers work.

    If IPC really is that important, it should be a default installed package regardless of whether any one program uses it or not.
    That is the big arguement. Even kdbus is an optional load module and being designed that way. Like something running apache webserver providing a web site may not need dbus/kdbus style IPC at all.

    The problem you have here is if you don’t need something and you have it anyhow it can cause security problems ssorbom.

    Yes address every single usage case option end up with a super big headache of conflicting interests. Desktop you require dbus. Some server operations you require dbus. Some embedded and server setups you don’t want dbus.

    IPC is important but there is more than 1 kind of IPC.

    dbus/kdbus are in the equal to Local Procedure Call under windows(yes something programs under windows don’t use much of) without the Windows size limits of messages. Due to the larger messages dbus allows the Linux desktop is being using dbus in may roles that com/activex would be under Windows. The key feature of dbus and Local Procedure Call is identifying the exact process the message came from.

    Yes we don’t even have straight 1 to 1 mapping between IPC’s between OS solutions.

    memfd + sealing is for file based IPC. Except this is Linux memfd is a file handle that exists as only memory data and it provides enough information to the scheduler to timeslice switch..

    http://andreaswarnke.de/ipc-quartztime/html/index.html You still find stuff like this that thinks that the issues are solveable in userspace without working in cooperating with the scheduler. These kinds of solutions data back to 1978.

    Something like ipc-quatztime on a server end users most likely will never notice a performance problem. Same solution used to move sound or video around you will notice the issues. Turns out desktop has a uniqueish requirement of having to work hand in hand with the scheduler.

  10. ssorbom says:

    But why isn’t dbus included in the debian-base meta-package? If IPC really is that important, it should be a default installed package regardless of whether any one program uses it or not.

  11. oiaohm says:

    tasksel is a program inside debian ssorbom.
    https://wiki.debian.org/tasksel
    Its there to provide you base system configuration options. As Robert did going out side these is untested territory.

    Robert
    Some folks just don’t need/want dbus on their systems.
    This here is a on going problem for systemd. The result has been systemd being forced to include a non dbus mode. Since systemd include a non dbus mode its not longer itself dependant on dbus. Logind on the other hand no dbus trouble but systemd will go forwards with no logind. Then you get some part like lightdm that demards to talk to logind you are now screwed.

    Systemd attempt to service everyone opens up these Pandora boxes because its willing to give up features that in a lot of ways should not be.

    Robert Pogson systemd itself is fairly straight forwards. When you have systemd mixed with sysvinit and people not wanting dbus things get confusing.

    Something really bad “service lightdm start” would not have made sure dbus was running yes this is sysvinit where ‘systemctl start lightdm.service’ if the service file was built correctly would have. So yes using the old sysvinit commands instead of the new systemd commands are exposing yourself to problems you would not see otherwise. Do these problems exist on prior sysvinit systems the answer is yes. Before logind the dbus issue started when you logged into your graphical environment and items started malfunctioning.

    XFCE4 is also dbus dependant. So if lightdm has run and you got into xfce4 without dbus running XFCE would be busted in places as well. This is why kdbus is becoming such a focus. The desktop needs a functional IPC. Preferred one you cannot fail to start without intentionally going out you way. kdbus for it to fail to start you have to blacklist the module and this will result it appearing in logging as application attempt to run it when its blacklisted.

    I think my way is better on older PCs of which I have a lot.
    Robert you way could be better but we really do need a better min tasksel also we need packages to be more careful to include everything in their dependencies and not presume that standard is installed or that systemd/init system is going to pull in that feature since due to people wanting to cut stuff down their is a really good chance it will not.

  12. ssorbom says:

    Some folks just don’t need/want dbus on their systems.
    But if it is being split into the kernel, its is coming anyway. I would be suprised if XFCE doesn’t use it. Most modern DE’s do.

  13. oiaohm wrote, “What robert attempted is an area where the Debian automated testing does not test.”

    This bug was reported months ago and the systemd changeover is so complex they could not deal with it easily, by, say, adding a dependency on dbus or such. Some folks just don’t need/want dbus on their systems. That said, it is a corner case that most users would not encounter if they want a desktop system. I like rolling my own, while others will accept reasonable defaults. I think my way is better on older PCs of which I have a lot. On a newer machine, it probably matters not a whit. Debian tries to please everyone but with systemd that’s not possible. I’m just thankful, I still have the option of installing and removing systemd if I want. If they take that option away, they’d better have systemd working smoothly. That might be possible in the next two years. My perfect world would replace systemd with something much more straightforward.

  14. ssorbom says:

    Grr… Typo. I was asking about the “tasksel” part

  15. ssorbom says:

    I meant, what did you mean by “Next only options are tested”?

  16. oiaohm says:

    https://piuparts.debian.org/
    Ssorbom I have said something in debian short language. QA is quality assurance. Debian has a lot of automated functionality testing. What robert attempted is an area where the Debian automated testing does not test.

    Quality of debian is part dependant on humans part dependant on the QA servers running 24/7 over every package change. There are ways to do stuff that is not covered by the Automated QA systems.

  17. ssorbom says:

    Next only tasksel options are tested out on the QA servers of debian. So yes what you did was going where QA has not been Robert.
    Huh? Would you mind rephrasing that? I don’t understand.

  18. oiaohm says:

    thr
    And I wouldn’t worry about systemd. Lennart Poettering devised PulseAudio, a colossal piece of crap that doesn’t work properly nine years later. Systemd surely is of the same quality.
    What if every idea here is all wrong.

    To be correct is is. PulseAudio could not work properly. Audio stacks in Linux have exactly the problem as video stacks. Yes the reason why X11 could not do tear free graphics meant Pulseaudio could not do perfect audio in every case.

    Issue is Linux kernel has Posix IPC. Posix IPC is a bit of crap. Posix IPC does not work with the sheduler. Remember Wayland and Mir the two targeted replacements to X11 don’t use Posix IPC but a new more cooperative IPC with the scheduler. The new IPC that Wayland uses is memfd and File Sealing. The File Sealing is fairly new and it tells the kernel when a process is done with the data and that the other end need that information now. There is no way in Posix IPC to say hey kernel this data I have put here into the Posix IPC should now be delivered to the other end now taking some of my processes time slice.

    The reality here is Lennart Poettering had a impossible task. He started working on systemd to find and fix why his task was impossible.

    The IPC is one of the biggest reasons why Linux has not been readly for the desktop. Poor grade IPC there is no way todo good quality audio or graphics without droping to using real-time modes that are highly power consuming.

    Basically there has been no power effective way todo tear free graphics and crackle free audio on Linux until memfd with sealing was added in 2014.

  19. oiaohm says:

    If that’s “correct” why do /etc/init.d/stuff still exist? Either I or Debian are confused, perhaps both.
    Robert Pogson I can answer this and you are not going to like the answer.

    Yes Debian is confused. /etc/init.d/ stuff has to remain for FHS compatibility. Yes sysvinit style of stuff was written into standard. http://www.pathname.com/fhs/ yes the last time that was altered was over 10 years ago.

    ‘systemctl start lightdm.service’ would have been the correct way. The lightdm old sysvinit script could have detect the presence of systemd and changed it code or the service command could have been replaced with a more intelligent one that redirects to systemd unit files if they exist. But due to debian wanting to maintain compatibility with multi init systems at the same time it gets messy.

    Next only tasksel options are tested out on the QA servers of debian. So yes what you did was going where QA has not been Robert.

    Logind requires dbus to work correctly and lightdm uses dbus to talk to logind. Horror of horror systemd will in fact work without dbus but items like lightdm are DOA without dbus in a systemd environment.

    Yes solid and secure IPC missing from the Linux kernel has lead to the hell that a user space service has to up and running. Posix IPC does not include and user identification information you can trust.

    I do agree the standard suite by tasksel could be trimmed.

    Systemd on debian is still running heavy. Due to the number of sysv script that the generator coverts to service files into a ram drive. A clean systemd is lighter. Also some of the problem is that the systemd .service files have been written in extremely min ways. Yes convert from sysvinit to service they just coded in call the old sysvinit file instead of redoing a proper service file.

    Systemd on debian is heavy due to how its been implemented. Arch and Fedora behave a lot better due to more .service files done properly.

    If systemd needs them, they should be pulled in by dependencies through APT. The tangled web that is systemd is still not straightened out.
    The problem you have to get systemd did not pull them because to systemd they are optional parts. lightdm should have pulled in dbus in a systemd environment because its not going to work without it.

    When it gets to interrelationships between packages with optional features things get complex fairly quickly. Yes some of the other DM would not have cared and if you were using lightdm with sysvinit and you missed console suite it would have died as well.

    There has been a stupid arguement against adding custom set dependencies based on what init system is installed inside debian. The result of people refusing todo this is what just happened to you.

  20. kurkosdr says:

    Lennart Poettering devised PulseAudio, a colossal piece of crap that doesn’t work properly nine years later. Systemd surely is of the same quality.

    So much about Freedum!! Who makes those decisions on what to include and what to not include anyway? Yes, I know theoretically the distro does, but practically, distros need to maintain compatibility with TOLD – That Other Linux Distro (tee hee), if they don’t want to be perceived as “hardcore”.

    So, unless you are using a “hardcore” distro (Slackware – yes) with less software support than the other distros, you basically play by Canonical’s decisions. Yes, that clueless, Steve Jobs wannabe space tourist owns all of you. If he develops a weird appetite for some Poettering’s festering pile of poor software engineering, boy will you have a taste of it.

    But you are free to examine, modify, run and redistribute distros that don’t contain PoetteringCode, as long as you have no problem being the 1% of an already small platform.

    PS: Sometimes, when I dream, I dream of an OS that contains little more than a protected task-switcer, network packeteer, filesystem and file browser, and every other functionality is built into the app.

  21. ssorbom says:

    Ohiaom and others here have already mentioned numerous benefits before. A slight increase in overall memory footprint is a small price to pay in exchange. I’ve looked at init scripts before, and trust me, systemd unit files make things much easier for packagers. I’m glad I’ve never needed to write one.

    I’m sorry you have had so many problems with it, but in the long run it makes for fewer bugs to not use init scripts. Look at the glass half full instead: if a bug crops up in systemd, it only needs to be fixed once, not a million times for a million different scripts.

    I rather wonder how you upgraded, because I installed a minimal system a few months ago and didn’t have any issues

  22. ssorbom wrote, “I think using using ‘systemctl start lightdm.service’ is the correct way to start that now”.

    If that’s “correct” why do /etc/init.d/stuff still exist? Either I or Debian are confused, perhaps both. I’ve been using Debian since about 2006 this way. I need a reason to change. systemd does not motivate me. I hope Debian figures things out. The transition was done just about the worst way possible, partly because of the many choices not making everyone happy and Poettering figuring he knows best… Testing systemd was the right thing to do. Making it the default for installations/stable was not. It should have been left in “experimental” until it worked smoothly. The fact that a choice had to be made was Poettering’s fault. He designed the damned thing. The fact that there is only one “testing” channel is perhaps a fault of Debian. One day we may all look back at this as the best/worst of times but I’m afraid Debian GNU/Linux is being made to function like TOOS, something I abhor. M$ made those kinds of mistakes in 199x. Why is Debian making them again in 200x?

  23. DrLoser says:

    I do say Robert, 5.6 megs of memory is most impressive! It is a shame that Microsoft cannot begin to match the capabilities of Linux, its sad really.

    Speaking of which, did you see this article?

    http://linux.softpedia.com/blog/Scientists-Use-Ubuntu-to-Interpret-Hubble-Telescope-Data-479321.shtml

    Seems that Linux is persuasive even when it comes to situations with large budgets such as NASA, the European Space Agency, CERN, Fermilab and the JPL.

  24. DrLoser wrote, “Who needs “basic utilities?””

    If systemd needs them, they should be pulled in by dependencies through APT. The tangled web that is systemd is still not straightened out.

  25. DrLoser wrote, “28MB virtual memory? “

    init on Beast takes only 1.4MB resident and 15MB virtual. Folks were saying systemd was more efficient. I don’t see it.

  26. ssorbom says:

    By the way, I think using using ‘systemctl start lightdm.service’ is the correct way to start that now. I would be a little curious to know if that would yield different results. I don’t think it is supposed to, but you never know…

  27. ssorbom says:

    Mr. Pogson,
    I had to re-read the specs you mentioned twice before I realized you were talking about being able to run a Desktop Environment inside of 5.6 megs of physical ram. That’s actually pretty impressive! Incidentally, what was it before? I haven’t seen those kinds of numbers for using a GUI since I ran Windows 98.

  28. DrLoser says:

    systemd, with a tiny GUI and TuxCommander running, uses 5.6MB RAM with 28MB virtual memory. 77 tasks running… Sigh…

    Ah well, it’s a Sunday and not a Friday. And you are Pogson and not Dolding.

    Nevertheless. 28MB virtual memory? You could factor that up by ten and it still wouldn’t be anything worth mentioning.

    BWAHAHAHAHAHAHAHA!, Robert.

    BWAHAHAHAHAHAHAHA!

  29. DrLoser says:

    Perhaps they need to redefine “minimal” by taking the basic utilies off the list.

    As if anybody would notice.

  30. DrLoser says:

    No, wait, you want advanced utilities, don’t you, Robert?

    Good luck with that. You can dismiss FLOSS “basic utilities” should you so choose.

    Advanced utilities require paying the vig to somebody like Red Hat, who will support your operating system of choice …
    … at a cost.

  31. DrLoser says:

    And here we see the power of the 1913 edition of Websters in action (my apologies, I had to resort to the online Merriam version, with added online power!):

    Utility, plural util·i·ties
    Full Definition of UTILITY
    1: fitness for some purpose or worth to some end
    2: something useful or designed for use

    Perhaps they need to redefine “minimal” by taking the basic utilities off the list.

    Boffo selling point there, Robert. Who needs “basic utilities?”

  32. DrLoser says:

    Still, it’s free, and it only took ten minutes.
    Isn’t IT a wonderful thing, if you don’t have any particular need to use it?

  33. DrLoser says:

    My first boot was a shock, however. I could not even log in thanks to systemd being unhappy (failed to start logind). I had to reboot to “rescue” mode to get in.

    Wot, no re-re-reboots?

    Good lord, Robert. This version of Debian isn’t even able to mimic Windows 95, is it?

    Never mind, give it time.

  34. thr says:

    Robert, that looks rad! Like Windows 95. And a Norton Commander/Speed Commander/Total Commander clone to boot. Is there something the FLOSS crowd won’t just shamelessly copy from successful proprietary software?

    And I wouldn’t worry about systemd. Lennart Poettering devised PulseAudio, a colossal piece of crap that doesn’t work properly nine years later. Systemd surely is of the same quality.

  35. DrLoser says:

    Well, no sane system needs a login daemon in the first place, Robert.
    It’s all good.
    Chuckle!

Leave a Reply