[rescue] Linux wet paint, was Re: Spark10 CPU question (must fix - SPARC damnit :-) )

[Stefan Skoglund (lokal användare)]

tor 2016-12-22 klockan 19:27 +0000 skrev Liam Proven: 
> On 18 December 2016 at 14:05, Rich Kulawiec <rsk at gsp.org> wrote:
> > Yes.  I use it as part of the interview process: those who have achieved
> > sufficient technical maturity and understand the Software Tools approach
> > as well as the fundamentals of the Unix philosophy can evaluate -- and
> > dismiss -- systemd in just a few minutes.  It is obvious on inspection.
> 
> Could you spell it out for me? I am trying to fully understand this.
> 
> The main caveats I'm aware of are:
> 
> * a large, complex process as PID 1 is undesirable
> * logging that is not to a plain text file is undesirable
> 
> I appreciate both, but OTOH both bring advantages too.
> 
> > Those who haven't are enchanted by its many bells and whistles, unable
> > to realize that the best possible future development path for systemd
> > would be to throw it away and learn from its mistakes.
> 
> And yet every major distro has now switched to it. I can't remember
> such unanimous uptake of a controversial new package in a decade, and
> only 2-3 times in the last 25y.
> 

A number of development/application stacks has started to depend on
systemd functionality so more work for a packager when including
upstream patches to apps/system software.

> libc5/glibc was contentious but everyone moved across in the end.
> 
> GNOME 2 won over everyone except some outliers, but GNOME 3 destroyed that.
> 
> I can't think of anything else,  not at system level. And even GNOME
> barely qualifies as that!
> 
> It's not because it's Red Hat. RPM is nowhere near so popular, for instance.
> 
> It *must* by simple deductive logic have strong arguments going for it
> or not everyone would be leaping aboard the boat.
> 
> And yet, almost all I read and hear is hate hate hate, with few actual
> specifics.

I personally dislikes Lennart Poetterings posture of network time
synchronization. Systemd's NTP support is SNTP which is a fair simpler
protocol/design.

You need full NTP support if the machine is to sync many machines -
timesyndc is one source only and it is also client only (and no hardware
devices so no GPS clocks.)

I think they should be a bit more honest about when SNTP and so systemd
isn't good enough. NO, i wouldn't run SNTP on a server - that requires i
think sync against multiple sources.