A friend of mine in another company asked me to post the following mail to
the list...

"We have a 997 5way HP3000 running one of our main applications. Back in
February this year, we began to experience Image database corruption
problems intermittently. It typically manifested itself in such as the
following...

ABORT:  DBDELETE ON DATABASE <<name removed>>;
TURBOIMAGE/XL ABORTS AT PROCEDURE: $00000198; ADDRESS: $00f35b38
TURBOIMAGE/XL ABORTS ON DBB CONTROL BLOCK
UNABLE TO LOCATE PRIMARY FOR PATH #3, DATA SET #100.


As this occurred again, HP became involved and made various suggestions to
correct the problem, including disabling DDX. They also suggested that we
enabled Image logging. We did that, but even when it did re-occur, HP were
unable to identify from the logfiles what caused the corruption.

If corruption occurred during the overnight batch, we would restore the
application (from a point before the corruption occurred) onto our
development box (996/600) in order to re-run the overnight batch and watch
it fail. We never got it to fail in the development box.

Over time, HP brought in a Disaster Recovery 997 box for us to run our
production on. For a while, no data corruption occurred, but then 2weeks
ago, it happened on the DR box.

At about that time HP said that a customer in the states, also with a 997
had experienced a similar problem. HP described the problem thus...

At periods of particularly high CPU utilisation, it is possible for a 64byte
cache area to not flush correctly. They also said that this problem is
peculiar to the 997 architecture.

They have now given us a patch for the problem which went into production at
the weekend, and so far, no new occurrences of the problem have occurred.

Has anybody else experienced a similar problem? If you have 997's do you
need to be worried, or need this patch that was given to us."

Mark W.
SPE.