The Story Behind ‘init’ and ‘systemd’: Why ‘init’ Needed to be Replaced with ‘systemd’ in Linux

I am subscribed to several mailing lists related to various Linux Distributions and Applications just to keep myself updated with what’s going on where. What are the new bugs? What are the Patches Released? What is expected in next release? and a whole lot of other stuffs. These days the mailing list is heavily populated with “Choose your side on Linux Divide”, mainly on Debian Mailing list along with a few other.

Linux Systemd
systemd replaces init
What “Choose your side on Linux Divide” is all about?

The init daemon is going to be replaced with daemon systemd on some of the Linux Distributions, while a lot of them have already implemented it. This is/will be creating a huge gap between traditional Unix/Linux Guard and New Linux Guard – programmers and System Admins.

In this article, we will discuss and solve following all queries one-by-one.

  1. What init is?
  2. What is systemd?
  3. Why init needed to be replaced?
  4. What features systemd will own.

What is init?

In Linux, init is a abbreviation for Initialization. The init is a daemon process which starts as soon as the computer starts and continue running till, it is shutdown. In-fact init is the first process that starts when a computer boots, making it the parent of all other running processes directly or indirectly and hence typically it is assigned “pid=1“.

If somehow init daemon could not start, no process will be started and the system will reach a stage called “Kernel Panic“. init is most commonly referred to as System V init. System V is the first commercial UNIX Operating System designed and usages of init on most of the Linux Distribution of today is identical with System V OS with a few exception like Slackware using BSD-style and Gentoo using custom init.

The need to replace init with something more perfect was felt from a long time and several alternatives were developed from time-to-time, some of which became distribution’s native init replacement, some of which are:

  1. Upstart – A init replacement daemon implemented in Ubuntu GNU/Linux and designed to start process asynchronously.
  2. Epoch – A init replacement daemon built around simplicity and service management, designed to start process single-threaded.
  3. Mudar – A init replacement daemon written in Python, implemented on Pardus GNU/Linux and designed to start process asynchronously.
  4. systemd – A init replacement daemon designed to start process in parallel, implemented in a number of standard distribution – Fedora, OpenSuSE, Arch, RHEL, CentOS, etc.

What is systemd?

A systemd is a System Management Daemon named with UNIX convention to add ‘d‘ at the end of daemon. So, that they can be easily recognized. Initially it was released under GNU General Public License, but now the releases are made under GNU Lesser General Public License. Similar to init, systemd is the parent of all other processes directly or indirectly and is the first process that starts at boot hence typically assigned a “pid=1“.

A systemd, may refer to all the packages, utilities and libraries around daemon. It was designed to overcome the shortcomings of init. It itself is a background processes which is designed to start processes in parallel, thus reducing the boot time and computational overhead. It has a lot other features as compared to init.

Why there was a need to replace init?

A init process starts serially i.e., one task starts only after the last task startup was successful and it was loaded in the memory. This often resulted into delayed and long booting time. However, systemd was not designed for speed but for getting the things done neatly which in turns avoid all the UN-necessary delay.

Features of systemd
  1. Clean, stateforward and efficient design.
  2. Simpler boot process.
  3. Concurrent and parallel processing at boot.
  4. Better API.
  5. Simple Unit Syntax.
  6. Ability to remove optional components.
  7. Low memory footprints.
  8. Improved technique to express dependencies.
  9. Initialization instruction written in config file and not in shell script.
  10. Make use of Unix Domain Socket.
  11. Job Scheduling using systemd Calendar Timers.
  12. Event Logging with journald.
  13. Choice of logging System events with systemd as well as syslog.
  14. Logs are stored in binary file.
  15. systemd state can be preserved to be called later in future.
  16. Track process using kernel’s cgroup and not PID.
  17. Users login managed by systemd-logind.
  18. Better integration with Gnome for interoperability.
Bottlenecks systemd
  1. Everything at one place.
  2. Not POSIX standard.

Systemd and Distro Integration

Linux Distribution Integration
Fedora Yes, first distro to adopt systemd
Arch Yes
RedHat Yes
CentOS Yes
Debian Yes, Debian 8 codename Jessie will have systemd by default
Gentoo Yes, but needs to be downloaded, installed and configure side with custom init
OpenSUSE Yes
Slack No (Though it has not been adopted till now in slackware, Patric Volkerding has not shown any indication if it will be adopted or not)
Ubuntu Yes, needs to be installed and configured with Upstream.
Controversy

Linus Torvalds, Chief architect of Linux kernel, feels attitude of key developer of systemd towards users and bug reports do not seems ok. It was also reported that systemd philosophy is weird and a foreign way to control system processes. The same has been recorded from Patric Volkerding and other notable Linux Users and Developers as well as over online forum, time-to-time.

systemd vs init

Features init systemd
DBus Dependency – Mandatory No Yes
Device based Activation No Yes
Device dependency configuration with udev No Yes
Timer based Activation Cron/at Proprietary
Quota Management No Yes
Automatic Service Dependency Handling No Yes
Kills users Process at logout No Yes
Swap Management No Yes
SELinux integration No Yes
Support for Encrypted HDD No Yes
Static kernle module loading No Yes
GUI No Yes
List all the child processes No Yes
Sysv compatible Yes Yes
Interactive booting No Yes
Portable to non x86 Yes No
Adopted on Several Distro Several Distro
Parallel service startup No Yes
Resource limit per service No Yes
Easy extensible startup script Yes No
Separate Code and Configuration File Yes No
Automatic dependency calculation No Yes
Verbose debug Yes No
Version N/A V44+
Size 560 KB N/A
Number of Files 75 files 900 files + glib + DBus
Lines of code – LOC 15000 (Approx) 224000 (Approx) (inc Codes, comments and white space) 125000 (Approx) (acctual code)

Conclusion

Anything running as pid=1 must not break, must not be mess and must be controlled by users effectively and efficiently. Many-a-user believes that replacing init for systemd is nothing more than reinventing the wheel everytime as a side effect of Linux. But this is the diverse nature of Linux. This is because Linux is that much powerful. Change is good and we must appreciate it if it is for a good reason.

That’s all for now. I’ll be here again with another Interesting article you people will love to read. Till then stay tuned and connected to Tecmint. Don’t forget to provide us with your valuable feedback in the comments below.

Avishek
A Passionate GNU/Linux Enthusiast and Software Developer with over a decade in the field of Linux and Open Source technologies.

Each tutorial at TecMint is created by a team of experienced Linux system administrators so that it meets our high-quality standards.

Join the TecMint Weekly Newsletter (More Than 156,129 Linux Enthusiasts Have Subscribed)
Was this article helpful? Please add a comment or buy me a coffee to show your appreciation.

73 thoughts on “The Story Behind ‘init’ and ‘systemd’: Why ‘init’ Needed to be Replaced with ‘systemd’ in Linux”

  1. Please do not refer to the Linux operating system as “GNU/Linux”. That is an inane sobriquet created in bad faith by troublemaker Stallman and should not be uttered by reasonable people.

    Reply
    • @Art,

      Thank you for sharing your perspective. I understand that the terminology around Linux can be a point of debate. I’ll ensure to use the term ‘Linux‘ in future references.

      If you have any more insights or preferences, feel free to let me know. Your feedback is appreciated!

      Reply
  2. SystemD is an init that wants to be a distro. It wants to control every process and every application with the distro. It is like a black hole that sucks in anything and everything that comes within its gravity field.

    SystemD violates big time one of the basic tenets of *nix – ‘Do one thing and do it well. Not only does systemd do hundreds of tasks (and wants more) but it does not do them that well.

    How can it be more efficient if, for every activity connected with it, we need a special program?

    If I did not know that systemD was developed by Leonard Poettering of Red Hat, I would swear that it was designed by Steve Balmer and developed by Microsoft to destroy Linux.

    Reply
    • I find it way too amusing, how you repeatedly say that Lennart Poettering developed systemd instead of MS. One year later the exact same guy (Leonard Poettering) works for MS.
      Coincidence? I think not.

      Reply
  3. Well, I’m calling it shytstemd, and here is why (I usually run Debian like distros) :
     
    * My laptop takes almost 4 minutes to boot where it took around 1′ with SysV,
     
    * It is a PITA to get rid of some of its parts (especially the network setup and its infamous resolved),
     
    * It _fails_ if you happen to have a fstab mount unavailable (SysV was just issuing a warning),
     
    * When you step backward to clearly see the whole picture, it is just like an octopus, it draws far too many things into its bed – this is the opposite of the way Unices in general and Linux, in particular, have always worked: the KISS principle,
     
    * Last but not the least: as it was adopted by almost all distros (wrongly and _very_ suspiciously because many of us, long time users, were dead set against it), that makes it a _very efficient_ trojan horse when the time will come to slaughter all non-PC opinions on the web, as we are already seeing this with GAFAM at this time, and my guts tell me that it is only a faint beginning.

    If you planned to do so, what a better option than to stuff the very process that controls everything in a machine and is so large that it makes it very hard to detect its flaws…

    Reply
  4. If systemd had been developed with a completely backwards compatible interface with the previous generation of commands such as service and chkconfig it would have been much easier to migrate to newer versions of Linux.

    If it were compatible but faster it would have eventually been accepted on its own merits. Throwing it into new versions along with a new firewall, new Satellite server, Ansible, etc just adds to the overall complexity of upgrading.

    I agree with this point: Clean, state forward, and efficient design. It may be clean and efficient but it is not a good thing to state forward; however, that imaginary word describes systemd in a most excellent way. I’m guessing that systemd was written and supplied to the public domain by Microsoft.

    Reply
    • Actually, it is the brain child (brain fart?) of Leonard Poettering of Red Hat. However, MS could not have done a better job of sabotaging Linux if they tried.

      Reply
      • Hi, at the (very dark) light of the past 3 years, I’d say with a quite high probability that it’s a big blocking/snooping hole P.O.S. – and I also remember having read some time ago that it is supposed to “revolutionize” the /home directory in a few versions from here :(((

        I think it’s time to make the move to Devuan – it stays behind Debian in terms of package novelty but it is a low price to pay not to use shytstemd anymore.

        Reply
  5. Note that the following paragraph has indeterminant meaning in English:

    “Linus Torvalds, a Chief Architect of Linux kernel, feels the attitude of the key developer of systemd towards users and bug reports do not seems ok. It was also reported that systemd philosophy is weird and a foreign way to control system processes. The same has been recorded from Patric Volkerding and other notable Linux Users and Developers as well as over online forum, time-to-time.”

    Reply
  6. Why there was a need to replace init?

    A init process starts serially i.e., one task starts only after the last task startup was successful and it was loaded in the memory. This often resulted into delayed and long booting time. However, systemd was not designed for speed..

    what the author should have said was: “However, init was not designed for speed…”

    Reply
    • No, the point was, “Systemd was designed to reduce the wastage of resources” like init.d. This helped speed, nevertheless it was not the objective.

      Reply
      • How can a systemd “reduce the wastage of resources” when it has 8 times the code lines and uses 12 times as many files?! Plus, like an octopus, it has its tentacles in most, if not all, applications.

        Reply
  7. that is the reason why i love Centos6 and had so much problems with Centos 7 !! I am not professional IT, but centos6 was very easy to configure my company network & workstations and centos 7 was hell.

    For example, wasted days to make extra license servers to boot up, after network card is booted, sometime concurrence is not needed, and saving few second at boot time is not a win. stability, predictability and easy control that what i need.

    Reply
  8. Jack, You don’t seem to have the knack for understanding Henry’s marketing strategy! His marketing strategy was make available to them what they need at a reasonable price and you will not have to convince them to want what they do not need. Not much different than Sam Walton’s. This is exactly the issue here. In the future your use of the indefinite pronoun may need refinement, unless you are an attorney, then shame on you!

    As a former hesitant Microsoft User and a current Linux user, for the last 5 years, all the people behind making Linux available deserve and have my lifetime support for their valiant effort. They have accomplished a unification of intent which I would have made any wager against prior to 1995.

    Lawrence October 12, 2016 at 5:30 am … welcome/ Innovation does not have to generate risk, just always insure your not getting what you do not want.

    Reply
  9. I find it interesting that this is posted by Editor? I guess it is easy to push unwanted changes on people if you act anonymously? Maybe Linux folks need to start looking at history before starting to change the environment that works just so they can make a name for themselves.

    I worked on real Unix system for decades. I used to consider Linux a Unix derivative. But I am doubting that based on the past few years. Linux is starting to act like Microsoft.

    Greg

    Reply
  10. Your history is a bit inaccurate. /sbin/init dates back at least to 3.2 release 4.2, and I’m told it dates back to System 3.1 that was fire-and-forget mailed as a digital tape to Berkeley that started BSD the *BSD variants.

    In fact, on 3.2r4.2 (Open Desktop), /etc/inittab was an aggregation of /etc/init.d/*, which is where the idea came from for Apache to read conf.d as a file-tree-walk in Apache-1.3.13 on-th-fly, which is now copied by everyone.

    /sbin/init was originally intended to restart printer services; it later was reused to restart vt100 terminal restarts when they died. The expansion to /etc/rc.d/ crap was done so that configs could be “added” as files when new software packages are installed (3.2r4.2 would “recompile” the config, svr5 — Unixware, Solaris — would read them in-place). Even using it to restart TCP services in general was seen as a stretch but /sbin/init performed well beyond its design scope.

    So, /sbin/init had very modest roots, was already performing beyond designed intents, but did its job very well. I’m not a fan of the systemd “do all the things” approach, but sometimes we need to make the bold step forward, try to iterate to improve, and see whether that gets us a benefit or a rollback.

    Reply
  11. sorry, but systemd is an abomination, and so is upstart, frankly. I don’t see why we replace the one thing that SysV got right.

    -mandrake (if you’ve been in the Linux world that long, yes, that one)

    Reply
  12. Systemd was a mistake to begin with. Change is coming (sinit, s6, openrc, shepherd, runit).
    https://devuan.org/os/debian-fork/

    You can add to systemd bottlenecks:
    bugs uncovered day in, day out (you remember about the “tweeter” thing? is that the kind of thing you would expect from /sbin/init?)

    when your root device fails, any already-opened session is rendered useless (can’t access systemd logs, can’t debug, can’t reboot, that’s something you used being able to do, when these used to be separate components, one might have broken without bringing down everything)

    AFAIU: there isn’t a single developer with a full visual on the whole project ramifications/understanding of what’s going on, what could go wrong.
    binary logs => logs corruption.

    Listing “low memory footprint” as a feature doesn’t make any sense: what are we comparing? have you looked at init?

    Saying systemd isn’t POSIX compliant is a cute euphemism. From day one: systemd didn’t comply with Unix philosophy. That should have been a red light, and I blame Red Hat here.

    Reply
  13. Why no one start something to opposite it and use init instead. I’m new to Linux but I agree with the term no broken don’t fix it. In fact a lot of people will be able to have a stable and effective system if the system stop changing dramatically.

    More contributors and newcomer like myself would enjoy Linux centos more. Please don’t end up like ms window os.

    Reply
    • Henry Ford said it best:

      – “If I’d of given the people what they wanted, they would’ve got faster horses”

      Same applies here.

      Reply
  14. “If you have to go along with the Ecosystem (here Linux Ecosystem) you should not say no!” If you keep resisting you will be left behind and the ecosystem will move ahead.

    This is the Microsoft attitude…a walled garden….it has no place in Free and Open Software. The objective of systemd is superimpose itself
    between the kernal and you and create a closed “ecosystem” closed off to the user and in control. We left such things when we stopped using
    Windows…..sorry.

    Reply
  15. systemd is good, its ok to have it. But imposing dependencies on other applications like DE(kde,gnome,cinnamon) etc is not good. I installed debian 8.4 today on virtualbox and removed systemd for sysvinit, i know that gnome3 has some dependencies(systemd) but i never knew that KDE plasma also depends on systemd, so i have no choice but to install systemd again and install kde though i didn’t remove sysVinit.

    Reply
  16. Change is not always good, whatever happened to “If it ain’t broke why fix it?” systemd is rogue-ware disguised as an open source alternative. The only thing stopping some of us from embracing FreeBSD fully is perhaps lack of KVM support. Linux feels like a M$ nightmare of modern times. People want freedom that’s why they come to the free world, not everyone expects they’ll be handed a box of hammers. At least the lawn is still green in the GNU landscape.

    Reply
  17. Already some time passed since this article. There is already systemd for ARM, at least. As for SysV not having dependencies when booting, Ubuntu and Debian already had some solutions for that. The core of question of the battle of systemd vs sysV init is mainly a relative newcomers saying we have got all wrong for twenty years, and ignoring our old axiom of using the tried, tested and stable…your table sums well some of the problems, including the complexity systemd brings unnecessarily to the table.

    Reply
  18. Hi avishek,

    nice post.
    I have still not blogged about this controversy in debian 8. being a big fan of debian i still don’t know whether to continue support debian or go with devuan (the new debian fork)

    Reply
    • Dear vicky,
      My simple Philosophy is – “If you have to go along with the Ecosystem (here Linux Ecosystem) you should not say no!” If you keep resisting you will be left behind and the ecosystem will move ahead.

      Also i never said, that change is always good.

      Reply
  19. Excellent article, but I’d like to suggest that you have someone whose native language is English proofread your work for grammatical errors. I’m not sure exactly what you meant to say here:

    “Linus Torvalds, Chief architect of Linux kernel, feels attitude of key developer of systemd towards users and bug reports do not seems ok”

    Reply
  20. Gluing many things in to system initialization program is not a good idea. This will lead to more vulnerabilities and undermine the UNIX (UNIX Clones) elegance . Hope systemd goes in a direction to get rid of things irrelevant to system initialization and process control.

    Reply
  21. You never go into the real problems of systemd, which is everything centralized. An init should only initialize, not trying to control everything. Plus, the fact that the developers tend to think they are Gods who can’t make mistakes, and blame everything on everyone else.

    One thing to do one jon and do it right.

    I honestly don’t care about systemd or any other init system, I just want to have the ability to chose whatever init I want. I don’t want to be forced to use a pice of shit. Even if that means that I wil have to stick with Debian 7 or change to another distro that don’t adopt that cancer.

    Reply
  22. A start job is running for ***

    I started with Red Hat in the late 90s. Tried a few others, ended up with Debian. Right now I am forced to use DHCP instead of static IPs because systemd hangs while booting machines with static network IPs because of yet another systemd related bug — somewhere.

    No more. I am done. PC-BSD goes on starting today.

    Reply
  23. Systemd is all about mission creep. It started as a replacement init, and is growing to encompass everything, the Poettering OS if you will.

    He actually said that he lied to further the golas of systemd, right after RH and a few others signed on.

    Systemd will evolve into Windows with a binary register before it isall done. One can only hope Slackware holds out, and that apps can be ported to non-systemd down the road. Otherwise, it is the death of Linux as we know it…

    Reply
  24. I don’t know much about SysV but I was pretty happy with OpenRC! Did you forget about that one?

    >DBus Dependency – Mandatory No Yes

    Well, depending on DBus pretty much sucks.

    >Timer based Activation Cron/at Proprietary
    I wouldn’t say Cron/at had the best ever user interface but those of us who were using it were used to it. I read that systemd can’t do random schedules like cron does. Do we really need to LOSE features? I liked that one!

    >Quota Management No Yes
    Couldn’t care less, no one else even uses my computer. But… filesystem quotas have been available forever and never needed special support from init. Is that no twhat is meant? What else would you put a quota on? Oxygen?

    >Device dependency configuration with udev No Yes
    Not sure what init has to do with this. Udev has handled devices and their dependencies for years now without systemd.

    >Automatic Service Dependency Handling No Yes
    I don’t know about your SysV strawman. OpenRC has service dependencies. It’s had them for years!

    >Swap Management No Yes
    The kernel has handled swap since before version 2.2 (where my Linux experience begins) It has never needed the init system to help with that!

    >SELinux integration No Yes
    SELinux has been around a very long time. Why does it need Systemd?

    >Kills users Process at logout No Yes
    The kernel does this. It always has. Why does it need the init system to take it over?

    >Support for Encrypted HDD No Yes
    There’s another one that has worked fine without systemd. It’s a kernel feature. WTF does init have to do with it?

    >Static kernle[sic] module loading No Yes
    What does init have to do with this? It used to be a kernel thing. I think udev does a lot of it now right? I know my Gentoo with openrc loads kernel modules all by it’s little old self whenever I plug in a USB device. Why do I need systemd getting it’s grubby fingers in there?

    >GUI No Yes
    I’ve tried various distros with boot-time GUIs. That is one feature I always end up turning off. I try to leave it on to make my system look more modern. Then I invariably get mad at it when I am changing hardware or otherwise messing with the system and I can’t see what is happening during boot.

    >Sysv compatible Yes Yes
    What does that even mean? If SysV compatibility is your goal then just run SysV!

    >Interactive booting No Yes
    OpenRC has an interactive mode. I can’t remember the last time I used it.

    >Portable to non x86 Yes No
    I don’t know about SysV but I find it hard to believe it hasn’t been ported to non x86. I’m sure OpenRC has. I know Gentoo has been available for other platforms for a long time. I don’t think they ship without an init system!

    >List all the child processes No Yes
    What?!? So systemd has top built into it? Why TF would I want that? It’s init! Everything is a child process of it! There have been umptine tools for listing/monitoring/managing processes in Linux/Unix forever! I don’t need the init system to also be it’s own administration tool!

    >Adopted on Several Distro Several Distro
    Something is very fishy about how quickly and without community input that happened…

    >Parallel service startup No Yes
    That’s been in OpenRC just about forever now! I keep reading that it is the main reason why we all need to switch to SystemD. WTF, I remember being excited about that feature when I was back at the apartment I moved out of in 2002!

    >Resource limit per service No Yes
    Just what every desktop needs! Were there really not already tools for those micro-managing SysAdmin types to do this already? I would have never looked at such a thing but I find it hard to believe that this is any newer of a feature than Parallel Service Startup.

    >Easy extensible startup script Yes No
    Why would I not want an easy extensible startup script. I even have single-purpose computers where I just put what I wanted it to do in inittab! It’s like an embedded device only much less work to set up!

    >Separate Code and Configuration File Yes No
    What? Code? Like in source code? I certainly never had to recompile any C to configure an init system. Is this refering to bash scripts? Bash scripts make pretty good configuration files! Normal users use the default settings anyway. Somewhat more advanced ones can edit them as easy as any other config file, the things you are likely to want to do are usually just name=value pairs, just like a non-script configuration file. But… if you really want to get fancy.. you can!

    >Automatic dependency calculation No Yes
    Works just fine in OpenRC!

    >Verbose debug Yes No
    Really? On SystemD I guess if you don’t see enough info to solve a problem just format, re-install and hope it gets better then?

    >Version N/A V44+
    Ok? No version numbers in SysV? Seems hard to believe. OpenRC is versioned, no problems there.

    >Device based Activation No Yes
    No idea! WTF is that? Do I need it?

    >Size 560 KB N/A
    560KB is nothing. How is this N/A? SystemD is just stored in the ether surounding the CPU?

    >Number of Files 75 files 900 files + glib + DBus
    And the winner is.. Not SystemD!

    >Lines of code – LOC 15000 (Approx) 224000 (Approx) (inc Codes, comments and white space) 125000 (Approx) (acctual code)
    That’s too many lines for an init replacement!

    Reply
  25. The whole problem with systemd is that it is trying to encompass functionality that lies outside the realm of the init system.

    The fact that a lot of key kernel developers are against it should speak volumes. Sadly it seems that a large portion of the Linux community nowadays consists of Windows converts, that do not truly understand the basic design of a *nix system. They want flashy bells and whistles like their former OS, rather than a well designed OS made up of tools that do ‘one’ job and do that job well.

    Reply
  26. Timer based Activation – Proprietary:-

    Now just why would that deadly word crop up in a vital section of a GNU/Linux installation?

    Proprietary to whom?

    Reply
  27. Yes, I would like an explanation of that this row in the table means, please?:-

    ‘Timer based Activation Cron/at Proprietary’

    especially the word ‘Proprietary’?

    Reply
  28. it would be nice to get more details or references to the specifics lited in the table.

    for example, as tracyanne asks, what exactly is this proprietary timer based activation?

    also, i wonder about a few more things:

    “Separate Code and Configuration File”: systemd does not do that? are you sure?
    i’d be inclined to think that it is the reverse. initscripts do mix code and configuration, while systemd’s config files are supposed to be declarative and hence contain no code.

    “Easy extensible startup script”: the only extra difficulty here would be having to create a script to be called by systemd instead of the actual program, whereas you could edit init scripts directly. does that make sysv init easier to extend? maybe, but not by much.

    lastly, interactive booting. i don’t know about systemd, but my sysvinit based desktop asks me at every boot if i want to switch to interactive mode, so that should be a yes for sysv.

    and like chris cox suggests, i don’t see many features here that could not be added to sysv init.

    i am with tracyanne. whatever. although i am excited about new features. i just wish we could have a more civilized discussion about it with more facts and less arguing for arguings sake. (i am referring to all the discussions on systemd as a whole here, not this particular article or comment thread)

    greetings, eMBee.

    greetings, eMBee.

    Reply
  29. Pathetic article and pathetic conclusion, systemd misses the whole point of unix , unix a herd of ‘simple’ and ‘focused’ daemons cooperating with each other to provide service. One giant daemon which does every thing ranging from device management to service initialization and what not . Systemd functionality needs to be divided in to multiple individual daemons which can be replaced by better ones if needed.

    Reply
  30. Hey guys,

    Why not also work towards developing a Linux Registry (a la Windows). This way we could get rid of that KISS way of doing things. Why keep things simple when we can over complicate things? Besides, this way we could also win the desktop. Who knows…

    Reply
  31. “with something more perfect ”

    This is an illogical statement.

    Perfection is an absolute.

    If something is perfect then there cannot be something else which is more perfect.

    However the problem is that in this earthly realm, all that is created by man will always be imperfect, since man is an imperfect creature.

    Reply
  32. Systemd is not just a Init system. There are big differences. Systemd more matches what I called a User-space Security System a long time ago as a way out the Linux Security Module fight.

    Linux kernel and BSD/Unix V kernel designs are major difference at process management. Linux and Solaris kernel have a closer design. Solaris runs something like systemd these days smf.

    The devil in the details. What is the problem. Its the Linux Kernel Process Management Design. Linux Kernel is incompadible with sysvinit. Upstart tried to use ptrace instead of cgroups to detect when processes broke away from patent process. Systemd used cgroups. The reality only cgroups work in the Linux kernel to process group so you can terminate them.

    BSD include process groups and the BSD init system wraps everything into what is called jobs.

    Richard Zimmerman and others who think BSD is some way to keep on using sysvinit need to go and install a BSD system and wake up BSD Init system is using process groups or the BSD equal to Linux cgroups. BSD process groups are functionally more limited than Linux Cgroups or Solaris Zones. Next problem due to the differences between BSD Process Groups and Cgroups is impossible to 1 to 1 the functionality.

    Not POSIX standard is a big one. Nothing in Posix standard defines functions or commands to place a application in a process group/cgroup/zone so that any break away application part can be terminated. Its not possible to write a generic Init system that works all the time while the API todo this does not exist.

    Sorry every one BSD system have never run a sysvinit system. BSD run a sysvinit like system with process groups. One of the reason why BSD has been a vastly more secure OS than Linux is the fact its services has been wrapped in process groups. The result of process groups is service stop equals the service absolutely stopped. cgroups in systemd openrc also result in the same thing so that stopped equals stopped. Please remember attacker breaking a service is going to attempt to fork away. Result with sysvinit like init systems on Linux was restart a breached service the attacker inserted backdoor could remain yet on BSD and Solaris this kind of attack would have been cleaned up by the Init system design.

    Like it or not we have to kill off sysvinit. Sysvinit is flawed. If any user decided to leave Linux and go to BSD because of systemd and end up using BSD Init atleast they have moved to a more secure and safer init system.

    The claim that sysvinit systems are working fine are in fact playing head in sand.

    Sometimes what is required for your own good is painful. Of course I don’t say we should not consider other options like openrc. Something you should not consider is sysvinit. Linux has kept sysvinit for way too long. Init system without something to group processes to make sure stopped is stopped is a bad init system.

    Reply
  33. @Richard Zimmerman: init might work fine in your case (I assume you mean sysvinit) but it certainly does not for everyone else, since every distro seems to be replacing it with ssytemd, or already replaced it with upstart, openrc, runit, etc.

    You are already being forced to use upstart if you use Ubuntu, so I don’t see what’s special about systemd that everybody complain about it.
    If you don’t want to use it, you also have Slackware (which I use) or Gentoo (which actually let’s you choose which init system to use) for example.

    @tracyanne: binary logs is a feature of the journald (part of systemd) but you can also use rsyslog or syslog-ng, in fact that’s how most distros I’ve seen use it.
    And the timer-based activation is not proprietary as in closed source, systemd is LGPL.
    I guess the author meant to say it was a particular feature or a unique implementation of systemd.

    Reply
  34. “Clean, stateforward and efficient design” – this is easily the most absurd claim I have ever read about systemd. On par with ‘war is peace.’

    Reply
  35. A quibble and a comment

    Quibble
    ++++++
    System V is the first commercial Unix, Sun Hardware came with SunOS a BSD derivative before System V was released. Xenix from (SCO/Microsoft) was based on System III as was AIX. NCR sold hardware called the Tower which ran version 7 long before even System III.

    Comment
    ++++++++

    Init is clumsy and slow but it works, literally millions of machines use it. Think of it like the standard transmission in a car. There are many better ways of doing it but this is the one everyone knows. I find it tiresome that I along with tens of thousands of other administrators have to relearn systemd when init isn’t actually broken.

    Reply
  36. Sometimes, it’s time to learn from the past and synthesize that knowledge into implementing something that is significantly better because of the benefits that it can provide for many. Systemd represents that leap forward. Instead of opposing it, I suggest that we learn to exploit systemd’s benefits.

    Reply
  37. I had initially installed systemd in my debian testing/unstable distribution. After having some time with it, I decided that I no longer want the damnable thing on my computer. So, yesterday, I replaced systemd with sysvinit. I still have some remnants of systemd, but I will have to remove about 300 packages, which I will do soon.
    But, for the most part, I am free from that systemd experience. If need be, I will do a backup and reinstall this week.
    So, in the end:
    Systemd been there, done that, removed it and got the T-shirt.

    Reply
  38. Excellent article, thanks. (*)

    > Logs are stored in binary file. and Timer based Activation == Proprietary

    Very well observed, tracyanne…

    I didn’t know about that proprietary aspect and cannot clearly imagine why it would be that way; I’ll try to investigate further, if nothing else to enlighten myself.

    Also, the situation has an air of “déjà vu”: it reminds me of the Bitkeeper saga and I wonder if things won’t end up the same way… like before, good people with the best intentions might get hurt in an unavoidable process.

    (*) O.T. (long)

    In my culture, we value a certain level of informality in treatment; formality is considered lack of courtesy and a way to express ill will against someone — or at least coldness. This is somewhat important for us.

    In other countries is the exact opposite, use of family name is considered more appropriate than addressing someone by the given name (this is viewed as some kind of invasion, it seems). Oh, well.

    Anyway, I’m so far from Asia that I cannot tell your name from the surname. And, as I said, using the complete name is not kind over here.

    We really should have some standard way to indicate which word is the family name, for instance.

    But thanks for an excellent write up on such an important theme.

    Reply
  39. Thank you. Informative article, but I’m a little lost in your grammar with this sentence:

    “Linus Torvalds, Chief architect of Linux kernel, feels attitude of key developer of systemd towards users and bug reports do not seems ok. It was also reported that systemd philosophy is wired and a foreign way to control system processes.”

    What exactly does Torvalds feel? Pro systemd? Con?

    Reply
    • I guess he has not opinion that he want to express than that one of the systemd developers was explicitly banned from adding anything to Linux kernel as they treated a bug in systemd as not an error of systemd and that Linux should be patched because of how systemd changed its behaviours.

      I guess that means that he isn’t that impressed in how they develop systemd.

      Reply
    • I think systemd supporters like to use SysV as a strawman. They can point out that SysV goes back to old Unix days while ignoring any updates which have happened along the way and completely ignoring other modern alternatives which were already widely adopted.

      Reply
  40. systemd doesn’t link against dbus. some of its components do.

    readelf -d /usr/lib/systemd/systemd | grep Shared
    0x0000000000000001 (NEEDED) Shared library: [libpam.so.0]
    0x0000000000000001 (NEEDED) Shared library: [libcap.so.2]
    0x0000000000000001 (NEEDED) Shared library: [libkmod.so.2]
    0x0000000000000001 (NEEDED) Shared library: [libseccomp.so.2]
    0x0000000000000001 (NEEDED) Shared library: [librt.so.1]
    0x0000000000000001 (NEEDED) Shared library: [libpthread.so.0]
    0x0000000000000001 (NEEDED) Shared library: [libc.so.6]
    0x0000000000000001 (NEEDED) Shared library: [ld-linux-x86-64.so.2]

    No dbus dependency in the main systemd daemon. Really stop making up false things about systemd for the sake of publishing articles.

    Reply
  41. I’m neither for nor against Systemd, whatever finally arrives in my preferred distribution is what I’ll use, I have neither the capability nor the time to quibble, and I do understand that change was needed, and yes init really is very slow to boot.

    But…

    Shouldn’t the following features of Systemd really be considered anti features

    Logs are stored in binary file. and Timer based Activation == Proprietary

    Reply
    • How can it be proprietary when systemd as umbrella is available under LGPL 2+ license? In addition, log file in text format is prone to attack from intruder who can easily tamper it without leaving a trace of presence.

      Reply
      • It could be proprietary if it isn’t documented.

        And about attacks? You mean like the one systemd did that made Linux kernel panic? It is noted and was what Linus T. wrote about.

        I prefere text log files as it is much easier to write and search in any way you like. Binary logs (and systemd) is the way of MS Windows…

        Reply
      • quote:: In addition, log file in text format is prone to attack from intruder who can easily tamper it without leaving a trace of presence.::quote

        And binary files can’t be tampered with??? In my experience that is exactly one of the major problems with the Windows Registry… it’s a binary file (several actually), and it is tampered with on a regular basis.

        Unless you are trying to tell us that because it’s binary no one but the Systemd developers know it’s format, that sounds like the old security by obscurity argument. Please let me introduce you to exhibit A… Microsoft Windows

        quote::How can it be proprietary when systemd as umbrella is available under LGPL 2+ license?::quote

        I haven’t the foggiest, read the article, that’s what it says in the article. Either it’s not proprietary and the article is incorrect, or it is, which makes it in my mind, an anti feature.

        Reply
  42. WHY do we need systemd? Init works just fine….

    I’ve not read one argument yet that states why I ahould be forced to use systemd.

    The answer is pretty simple, allow us to use either init or systemd.

    If that isn’t possible, I guess I’ll be looking into FreeBSD like others are…

    Richard

    Reply
    • SysVinit was on death list for a long time (about 30 years according to some UNIX administrators themselves). It is a relic from the old technologies that even UNIX systems ignore it in favour of their own init implementation because of bad concept such as mixing configuration as executable. Sysvinit has complex implementation for basic stuff it has become a burden. Look how Linux distributions felt short of needing a standard within SysV coming with their implementions. Systemd solved that problem.

      Nobody forces you to use a system running on systemd because your distribution chose it because of easier of maintenance. You are free to switch to non-systemd distribution.

      Note that FreeBSD is implementing launchd that inspired systemd and it does not use SysV.

      Reply

Leave a Reply to Akshat Cancel reply

Thank you for taking the time to share your thoughts with us. We appreciate your decision to leave a comment and value your contribution to the discussion. It's important to note that we moderate all comments in accordance with our comment policy to ensure a respectful and constructive conversation.

Rest assured that your email address will remain private and will not be published or shared with anyone. We prioritize the privacy and security of our users.