Bill Karwin wrote:

>
> Ann Harrison wrote:
> > It's a classic problem with all databases, I would think. Alisdair
> > is correct. The reason for not opening a database on a remote disk
> > is that we can't know whether the database is already open elsewhere.
> > That's true in the "classic" architecture because each attachment
> > opens the database file, so file sharing is essential. In the
> > superserver, we may be able to fudge that rule.
>
> Could another compromise be that the access method permits a database to
> be readied if it is on a remote filesystem, BUT sets the read-only
> database flag when it does so?
> I'm not suggesting that the access method change the flag in the
> database, only in its attachment to the database, to pretend that the
> read-only flag is set.

Bill, I like your above idea. InterBase servers from IB 6.0 onwards
already have the capability of querying read-only databases. Hence, they
could treat a NFS-mounted remote database as a 'read-only' database and
set this flag only in its runtime cache.

The access method needs to be informed about the client's intention
before hand though. This could either be done by introducing a new DPB
parameter or using an existing one. Alternatively, the local
'isc_config' can be (by adding new parameters) used to define all the
databases that could be accessed for 'RO' purposes on the NFS-mounted
fs's. I'd prefer the DPB option myself.

>
> Maybe a better way to handle that is to change ibserver so that it
> permits a ready of a remote database _only_ if the read-only flag in
> that database is already set. This would allow DBAs to place gdb files
> on mapped drives, if the users really need access to it through that
> connection.
>

In the above case, say the DBA wanted to make the RO database RW at some
point of time in the future. The advantage of having all the clients
accessing the database through a single server is that the server could
be informed to disallow any new connections so that the database could
be transformed to be a RW. In the case where it is shared on a file
system, there would be no single gateway by which all the servers would
be accessing the database. How would the remote servers be informed of
the DBA's intention to do the RO->RW transition?

Best wishes,
Sriram

> Bill Karwin
>
>
> To unsubscribe from this group, send an email to:
> IB-Architect-unsubscribe@onelist.com

--
Sriram Balasubramanian EMail: bsriram@...
InterBase R&D, Inprise Corporation

[Non-text portions of this message have been removed]