Dinamic Reconfiguration

I scripted something like this at one point on an E10k, to deal with a
seasonal processing requirement, but VCS was not conscious of it
directly through the API. The API state information isn't really
designed to deal with H/W information only with application level
interfaces so I just set it up as script that executed during failover.

As you've said don't get involved with any system board with IO on it.
AP makes a mess of the whole thing.

There were two additonal caveats I found:

1
Make sure your floater board(s) is the highest system board number on
both the source and destination domain. So for example

Domain A 1 3 5
Floater board 7
Domain B 2 4 6

Memory and CPU's utilization is appended from the lowest level PROC
number so your best chance for clean application quiescing and memory
drains is on the last board.

2

Understand how your appliction handles parallezation and what prompts it
to grab more processors. Architecturally DR really IS NOT designed to
be an unattended operation. The goal is to meet seasonal processing
requirements not HA.

After making it work I removed it for reason 2 and my gut feeling was
that it was heavily prone to pilot error since DR operations need to be
done with a DBA.

Regards

Javier Alonso wrote:
[color=blue]
> Could VCS support Dinamic Reconfiguration (only attach,
> never dettach and no IO involved , only CPUs and memory)
> in a E10K Solaris box.[/color]

--
Todd Kiker

desk: 760.809.2136
fax: 509.691.4286
[email]tkiker@iname.com[/email]