On Fri, 29 Oct 1999, Stuart Robertson wrote:

> I need to setup an ADSM system that will fail over to another RS/6000 if
> there is a problem with primary system.
>
> The system as it has been installed (not by me!) has the database and
> recovery log volume group failing over - which is obviously required.

I do not know what the experience has been at other customers: ours were
extremely bad. HACMP *caused* the most serious problem we ever had with
ADSM. Basically, the scenario was that network connections were failing
intermittently with a high frequency of down and up again, while the
systems remained fully operational. The result was that for some time both
systems seem to have worked on the same database and destroyed it so
thoroughly that it could not be recovered by an audit. IBM was not overly
interested in analyzing the problem at that time, so we never learned what
was really happening.

I am not saying that this is a necessary or even probable scenario to
occur. I *am* saying, however, that we have not found any indication that
ADSM is reckoning with the possibility that it will be taken over by a
different system while the database is open, so that I would still
estimate that the risk of causing damage by HACMP is higher than the
perspective of preventing damage. If IBM thinks otherwise, I in your place
would insist on a contract that explicitly guarantees consistence of the
database in case of failover, with high penalty and full recovery of all
losses and damages, should anything happen.

Best regards,

Helmut Richter

==============================================================
Dr. Helmut Richter                       Leibniz-Rechenzentrum
Tel:   +49-89-289-28785                  Barer Str. 21
Fax:   +49-89-2809460                    D-80333 Muenchen
Email: Helmut.Richter AT lrz-muenchen DOT de    Germany
==============================================================