That would all depend on the nature of the program. If the IO is small
fcontrol the critter. (Small meaning few, not size of records) Otherwise XM
would be good, thought XM handled most IO's. When you restarted the machine
did you tell it not to recover. My understanding of XM is that all
transactions done to a disc are buffered in a known area, disc log, and then
as time permits or space it is then sent to the respective areas on the
disc. Thus when the system goes done the XM then, looking into the log sees
what was done and not done and finishes the IO's.
But then again I could be wrong. Instead of attaching to XM, I thought you
had to do something special to detach from the XM to speed up your IO's with
the added risk of losing data. Programs which can be restarted without
much-a-do are ready candidates for this.

James

-----Original Message-----
From: HP-3000 Systems Discussion [mailto:[log in to unmask]]On
Behalf Of #Agnes Hajdu
Sent: Wednesday, November 17, 1999 1:34 AM
To: [log in to unmask]
Subject: Write directly to a file


To all,

After a system-crash we discovered that the content of one file was
lost. This probably happened, because the information was still in the
memory, while the machine went down.

I'm looking now for a possibility to prevent this to happen again. The
idea is, to force the write-command to write directly to the harddrive.
A got several tips from colleagues, one was the FCONTROL statement and
another a so-called transaction manager. Does anybody have informations
or experience with these features? Or does anybody have another idea?

Thanks
Agnes.