What I find incredibly interesting any time the concept of fragmentation comes up is how
did several versions of UNIX with slightly differing interfaces create such a headache for
UNIX vendors and users in the day, but now we've got a Linux/BSD landscape out there
with still pretty significant differences between distributions and UNIX's progeny
seem to be doing just fine.
Were users looking for different things from their computers in the 90s vs today? Have
folks just gotten more used to variability in computing environments and just accept it as
part of the plan?
What comes to mind for me is the different init systems, desktop environments, networking
tools, user management tools, and basically that anything that isn't lore in POSIX
seems to be up in the air these days. However, you go back to when SVR4 derivatives were
king, they all had the same init, the same useradd, the same /etc/passwd, the same
ifconfig. Maybe some of the snazzier new features were pretty variable, but the most
basic stuff like starting your system, creating a user, seeing if you were connected to a
network, essential administrative functions, were relatively consistent.
Nowadays I have to wonder if my init system is runlevel based or some systemd monstrosity.
I have to question whether I can rely on useradd or some other tool being present or if I
should forgo it all and just edit the /etc files directly. Heck, I couldn't say
which but I seem to recall a distro I played around with in the past year where this
actually didn't work, I had to research whatever arcane user management tools they
shipped with that one because whatever they chose broke with convention so much. I have
to pray it has ifconfig or else go look up the docs for iproute2 and iw because nobody can
make up their mind on what to replace ifconfig with, just that it has to go and replacing
it haphazardly and non-universally is better than fixing/modernizing it.
Not looking to start some great debate over which of these components is ideal of course,
just remarking at the fact that in the early 90s, if you were on a contemporary UNIX
system, you'd probably have no trouble modifying system init, adding users, networks,
etc., but today I sit down at an unknown Linux machine and I have no confidence that the
particular flavor of system administration that I'm used to will be even remotely
represented in the subset of tools that particular distro ships. Luckily, it's free,
so perhaps that is what has made the difference, folks are more willing to deal with
variability when they aren't paying for what should be a consistent experience, but
regardless, the fragmentation in Linux world today feels like it is much more severe than
UNIX was in the past, but that's also looking through a lens upon a time I certainly
wasn't cognizant of this stuff in.
Anywho, that was definitely an informative read, thanks for the share. As someone who is
constantly trying to dial in my own personal Linux distro, the questions of
standardization and uniformity feature in my mind often.
- Matt G.
------- Original Message -------
On Wednesday, November 9th, 2022 at 2:51 PM, Joseph Holsten
<joseph(a)josephholsten.com> wrote:
On Wed, Nov 9, 2022, at 14:16, Larry McVoy wrote:
On Thu, Nov 10, 2022 at 09:01:42AM +1100, steve
jenkin wrote:
I???ve only recently stumbled across this paper.
It gives the answer to one question I???ve had:
Why did Linux become more popular than everything that came before it?
Yeah, that was a difficult time. My boss, Ken Okin (SVP of all server
hardware) didn't like the switch from SunOS to SVR4 any more than I did.
He paid me to go argue with the execs for 6 months. That paper was
the result.
It obviously went nowhere and Linux won. Big surprise.
The one thing I learned in that 6 months was respect for the execs.
As an engineer, I had the luxury of taking the time to solve a problem and
know that I solved it correctly. The execs didn't have that. They had
to make decisions essentially with their gut, they couldn't afford the
time to figure out the right answer, they had to come up with the right
answer on the fly. I don't think I could do that.
It’s painful to look at where (Open)Solaris was when Oracle acquired it and where it is
now. SMF, Zones, ZFS, dtrace, mdb. Oracle Cloud doesn’t use Solaris for anything. I can’t
recall hearing anyone using dtrace or ZFS around the place.
Meanwhile, illumos derivs have actually done interesting things. Not that NexenStor or
SmartOS have made a big dent, but at least they’ve had more recent ideas to copy.
--
Joseph Holsten
http://josephholsten.com
mailto:joseph@josephholsten.com
tel:+1-360-927-7234