[GoLUG] "Who are we to think has greater knowledge on the topic?" "Me. Seriously" -- Steve Litt

[Barry Fishman]

On 2025-08-27 11:13:39 +01, kc-golug at chadwicks.me.uk wrote:
> 27 Aug 2025 09:36:38 Ron <ron at bclug.ca>:
> SysV was pretty bad for end users to easily use. Systemd isn't much
> better for end users but perhaps for developers. Runit is better for
> both, the best possible probably not but the best I have used on Linux
> which is pretty much just openrc, sysv, upstart, daemontools with
> qmail and systemd.
>
> I believe systemd was developed for cloud system usage. Upstart I
> guess for end users but not sure. Ubuntus CEO said he wanted to stick
> with Upstart but Debians decision (coerced decision) was the
> end. Upstart like systemd wasn't much better certainly for end users
> than sysv unfortunately. Debian switched to systemd because great
> efforts and lies like being much faster when ssds were the real reason
> or gnome packages depending on systemd had meant Debian devs had
> switched some things to systemd without an official decision. Better
> to ask forgiveness kind of thing. Well it worked and now it's a mess.

My memory of the time was a bit different, and my focus was on the
desktop situation.

There was concern that boot times were too long.  There was knowledge that
slow disk bandwidth was part of the problem.  But focus was also on the
fact that by building a directed graph of the component dependencies,
more components could be brought up in parallel.  The desktop would then
come up when all its required parts were available, and not have to wait
till all the servers were up and running.

True, people could hand tune there startup order as was done in Arch, or
rely on a fixed set of stages to control this, but this seemed harder to
setup in a general way.  It was thought that it would be better to
supply a means added as a part of each package, aware of its
dependencies, to create files to express these dependencies that all the
package managers could interpret in setting up their underlying systems.

For example Debian, Fedora and SuSE all used a System V style setup, but
used different boot stage setups and shell functions in building their
startup script, so each portable package had to be aware of all that,
and couldn't have a simple startup/shutdown script that fit into all the
distributions that used it.  Choosing their own independent way just make
boot logs much harder to read and process.

SystemD supplied a mechanism where each package could define its
startup/shutdown requirement in well defined files which could be
dropped in a common system directory and let the overall OS distribution
automatically (via SystemD) work out a boot/shutdown process aware of
how all these dependencies interacted.  As I recall it did demonstrate
faster build times for early users, but had some problems in shutting
down systems where certain components failed to close down properly, and
the system hung up waiting for them.  Rather than a quick "sync; sync;
sync; halt" Systemd wanted to handle servers that need time to save
their current memory state to a file, but any process which wanted to
wait for something, like all its active transactions to complete, or a
broken server could hold up the entire shutdown.  It still can, and does
for me, regardless of the SystemD based distribution.

Ubuntu developers had some issues with the direction SystemD was going
and came up with Upstart as an alternative approach that they felt was
better solution to the same problem.

My exposure to other init frameworks came after Debian settled on
primarily using SystemD, and many in its user communities rebelled.  Why
Debian chose SystemD is something am not familiar, but it seems it was
effected by the common "Free Desktop" development which was being used
by Gnome (and later KDE) and was internally using SystemD files to build
there complex package framework.  Porting this to other init setups
seemed to something they didn't want to maintain, although some other
distributions did do it.  Ubuntu did keep developing a parallel setup
with Upstart, but after a while felt it also was not worth the extra
effort.


As with Upstart, Ubuntu felt they had a better solution than Wayland to
as a window system, which they called Mir.  During the downsizing that
Ubuntu was going through at the time, Upstart, Mir, and the Mir based
Ubuntu Mobile Phone were dropped because they felt they could not afford
the cost of going their own way in these areas, and it was better to go
with supporting (and possibly influencing) more established approaches.

As I said these were my impressions at the time, and came from just reading
the mailing lists and using the distributions involved.

-- 
Barry Fishman