Andrew Brown writes:
 > >I have successfully used a similar method to Andrew Brown for dynamic
 > >OS upgrades - with a minor variation prior to the kernel build;
 > >
 > >  in the usr directory I replaced more directories;
 > >    lib libdata libexec bin sbin games include mdec obj share
 > 
 > well yes, of course :).  include and libdata need to be done at the
 > same time as well.  the others are somewhat less version critical.
 > 
 > >Once these were in place, the host (a firewall) continued running
 > >without a problem for the next 8 hours, until it was convenient to
 > >reboot with a new kernel.
 > 
 > ...but since you had installed the new kernel, the only thing
 > remaining was to actually tell the machine to reboot, yes?

Yes, a reboot was all that was needed - this was the only interruption
to supply. I wonder whether this procedure will be possible with a
more disparate upgrade - eg to 1.3. That is, will a 1.2 kernel run
with 1.3 tools for the duration of the kernel rebuild period. Of
course, once you have the first machine in each architecture upgraded,
it can be used to pre-build kernels for distribution, thereby
shortening the changeover period.

PS. One item I missed during the upgrade was a tcpd which we had 
installed in /usr/libexec - instead of /usr/local/libexec.
------------------------------------------------------------
Rex McMaster                       rmcm@ariel.unimelb.edu.au     
                                  rmcm@paradiso.apana.org.au