System Restore

[Eric S. Johansson]

Ante Karamatic wrote:
> If you used System Restore on Windows, you would know it's useless even
> there.

just because one instance has a bad implementation, it does not 
invalidate the entire concept.

> There are some inovations in Windows that are good, and thoose
> inovations will be implemented in Linux's window managers (window
> managers, NOT Linux). It's very important to understand that Linux, as
> kernel, sysv init and libc can't have system restore. It's pointless. It
> would be step back, because now you can have multiple kernels and
> multiple inits. If you loose libc, even system restore will not work,
> cause it needs C library.
> 
> People who think that linking whole /etc to something else don't have
> too much exp. in Linux or UNIX operating systems.

while I do admit I don't have too much experience with UNIX (20+ years) 
I do know something about designing systems and the human interface to 
those systems.

my proposal is for partitioning the system between the baseline and a 
site-specific changes.  The reason for this partitioning is twofold. 
First is to make it easier to isolate and manage local change so that if 
you need to reproduce it, you can with greater ease than you can today. 
  The second is so that you can automate much much more upgrades not 
only of executables but of configuration files.  If you can identify 
changes in the right way, you can eliminate the need for human 
involvement in the upgrade process.

my experience with gentoo has reinforced my desire to minimize human 
involvement in incremental system upgrades.  It is ludicrous to need 
someone to pay attention for hours to a system compiling a months worth 
of updates just in case something breaks.  And then, after it's all done 
to spend God knows how much time updating configuration files.  Now 
multiply this process by 10 or 15 systems.

Debian has the same problem to a lesser degree.  Fortunately, releasing 
binary images does make it easier to detect and manage dependency 
problems but configuration file changes especially when upgrading 
releases is difficult.

I believe it's possible to reduce human involvement on upgrades by at 
least one order of magnitude.  Which means easier upgrades for less 
technically astute users and system administrator talent can be spent on 
more important tasks.

---eric