Current strategy:

1. Plan an extended downtime (plan for 3 hours, hope 2 will suffice)
2. Apply patches
   1. 112186 - OBP - requires single-user mode and a reboot, so boot -r
   2. 111846 - cfgadm - requires a reconfigure after, so another boot -r
   3. after that will want to make sure system still works fine in multi-user mode
   4. 115275, 116970 - mpt, sbdp - both require single-user mode, so yet another boot -r
   5. 109892, 111412, 111413, 111095, 111096, 111097, 111847 - some of these have cross-dependencies on each other. Most require reboots, none require single-user mode. So apply these and then just boot -r.
3. If patches alone are insufficient, rebuild the arrays:
   1. take c1t4d0s* out of its metadevices - manually fail it in d3 and d8
   2. clear it from the metadbs
   3. luxadm to get rid of it and reconfigure it
   4. wipe and restore the partition table
   5. add metadbs back
   6. start recreating metadevices, starting with smallest first
      1. d3 - swap
      2. d8 - fsys1
      3. d5 - /u3

We can assume 15m per boot -r iteration, and we have 3. The patches are all of varying sizes, some will take just a few seconds to apply and some will require several minutes - assume 3m per patch. I've never flashed OBP in a v880 before; it took about 3-4 minutes in an Ultra1 and maybe the same for a Blade 1000. I can't expect it to be any faster for an 880, let's say 10m. Add some time for fudge factor too.

So maybe hoping for 2 hours is expecting a bit much. Still, 3h seems a reasonable estimate for a downtime, but maybe ask for 3.5. Would it be easier on the user community if we broke this up into a few separate downtimes? Keep in mind that once all this is done, rebuilding the arrays takes hours, although the system is still usable then.

-- MikePatterson - 05 May 2005