Conference iosg::all-in-1_v30

    
    Terry,
    
    What you are doing is very difficult to get to work from a FCS point
    of view and close to impossible (if not impossible) if you are running
    at distribution level 0.
    
    You have 2 problems here:
    
    1) ALL-IN-1 does not use the FCS for all FC functions. It uses it only
       when either a remote drawer is accessed, or a cross drawer 
       operation is done. It also used it for all IAD functions as you
       saw.  Your first problem is to convince ALL-IN-1 that the cluster
       is really not a cluster. I will leave that to Stuart or others
       to tell you if and how it can be done.
    
    2) You second problem is convincing the FCS that the cluster really
       contains multiple partitions.  Keep in mind that a "remote" drawer
       is simply a drawer in another partition that is not serviced by 
       the current server. The physical location does not really matter.
       There is a way to do this, but you need to run your servers at
       distribution level 1 and use DNS partitions.  This is how our
       tests work when testing brokering (some of them at least).
       If you are interested in this let me know, it is an undocumented
       thing so I would rather not put it in the notes files.
    
    --Bob
    

Bob,

The problem does not seem to be with the FCS. The two ALL-IN-1 systems are
completely seperate except for anything in SYS$COMMON. This means that both
use the same FCS image, but the OA$* logical names on each system are different
so the server on NODEA finds a different PARTITION.DAT and PARTITION_MASTER.DAT
than the one running on NODEB.

I turned tracing on and IAD went via the FCS, but when I tried to add the
drawer to the file cab (ADR) there was no FCS access. It appears that ALL-IN-1
somehow decided that the drawer was local (presumably because it was on a node
in the same cluster) and tried to access it directly. If the drawer happens
to exist on both ALL-IN-1 systems the add works but add the local drawer
instead of the remote one. If the drawer only exists on the remote system
the error is output.

I believe the problem is that ALL-IN-1 decides not to use the FCS under 
these circumstances, but I may be wrong of course.

I guess there is no way to change this behaviour, but just in case there was
I thought I would ask.

Thanks,

Terry

    Terry,
    
    there is no way to have more than one partition on a system as long
    as you run at distribution level 0. ALL-IN-1 (and the FCS too as far
    as I remember) will check the nodename of a drawer against the names
    of all cluster members. The effect is that ALL-IN-1 considers all your
    drawers as being local and expects to find them in the PARTITION.DAT
    it has access to. And the same problem would arise in the FCS if you 
    would pass one of the "remote" drawer names to the server through 
    a FILECAB function.
    
    Cheers,
    	helmut

Thanks Helmut.

I thought that would be the case, but just in case there was an undocumented way
of getting round it I thought I'd ask.

Regards,

Terry