>In-place upgrades do a huge amount of space checking since a failure due
>to in sufficient space would essentially render dead system.
> These assumptions in the current code are pervasive and complex.
>Live Upgrade by-passes this code (albeit in a rather ugly manner).  If LU was 
>the
>only way to upgrade, then the simplification around this would be huge. 


But the space checking has always been rather simplistic; there was no
or little relation between the projected space free after the upgrade or
the actual space free after the upgrade.

On several occasions I had to remove things before upgrade only to
install them again afterwards and find that there was a GB or more free.

Some of the checks could be very simple; just require a few GB of free 
space and you are guaranteed it will fit.

>I am assuming you are saying this because of limited disk space ?
>On x86 systems, a full install takes about 3.3GB.  Double that it's 6GB.  
>So what size disks are on these systems ?


I've found few roots < 5GB; but the problem with x86 disks is often:

        - I have Windows, Solaris and Linux
        - I have set aside XGB for Solaris and now I suddenly need 5GB
          more.

(With the current liveupgrade scheme it s much more but with liveupgrade
on ZFS it does seem to be a lot less).



I still don't like forcing ZFS on people, though; I've found that ZFS
does not work on 1GB SPARC systems; I found that a rather high lower limit.

(Whenever the NFS find runs over the zpool, the system hangs)

Or is making Solaris w/ ZFS run in 256MB a goal or requirement for
Caiman install?


Casper


Reply via email to