On Sat, Jul 24, 2010 at 11:50 PM, Stefan Parvu wrote:
> IPS tries to be a network pkg management system. It needs to be simple
> ... list of good ideas ...
These are all things that could easily be addressed, and aren't per-se
reasons to abandon IPS. If it was important, these things could be done.
With the new on-disk format, I'd assume it would be easier to maintain a
mirror using the existing mirroring tools...
As Machiavelli said, change is opposed by those with a vested interest in
the status quo...
> What comparative analysis have you done
Last time I looked, probably 5 years ago, and if I recall correctly, the
basic work flow in the SVr4 package tools was something like
start pkgadd, pkgrm, pkginfo, ... process
read /var/sadm/install/contents into memory
install/remove/whatever a package
read /var/sadm/install/contents and merge/modify the lines into a new
/var/sadm/install/contents file if the world was modified
exit process
for every single package installed. The thrashing over the
/var/sadm/install/contents file was more than huge, and is one reason why
patch management in zones sucked rocks so badly (tens of hours to days to
upgrade a system with many zones...). There were some simple comp-sci style
modifications that could/did speed things up by orders of magnitude, but
they required replacing the /var/sadm/install/contents text file with some
sort of structured object (database, btree, whatever) that was incompatible
with way too many things that had come to depend on the project private
contents file. There were proposals to cache the file in memory (ala nscd),
to make it a special kernel driver (ala /etc/mnttab) and others, none of
which (IIRC) made it past the proposal/prototype stage.
-John
_______________________________________________
opensolaris-discuss mailing list
[email protected]
