RE: ODBC (was RE: Green screen - it's time is over )

Your client calls a server that returns data in a message.  The message
layout may or may not reflect the current physical layout of the data on the
disk.  If the physical layout of the data does change, then the server
simply maps that new layout onto the original message.  The client code does
not change.

Note that this is not referring to the situation where you have to actually
add or remove LOGICAL elements of your database - in those cases, of course
you would have to change your messages as well.  Instead, I'm talking about
the more mundane task of renaming files or fields, moving data to different
libraries, moving fields from one file to another, and so on.  Or the
sligntly more exotic situation of merging a foreign database in with your
existing application, as in a corporate merger (I've found this to be one of
the situations that benefits most from a distributed architecture).

This ability to change the underlying physical layout (including, for
example, the fact that a file is physically moved to another machine, or
even another data format entirely) is the primary reason that a
message-based architecture is so preferable to ODBC.

This same argument doesn't apply ODBC/SP, where ODBC is used to call a
stored procedure that dynamically builds the result set.  That is logically
equivalent to a message-based design.  And for situations where the message
only has a single layout, you can even use a program call rather than a
server, although servers have many other benefits that may outweigh the
simplicity of the program call interface.

In any event, with a server, I can change the table name and the column
names (or, as I pointed out, the machine that the data is located on, or
even the actual data representation, such as from EBCDIC to ASCII).  All I
have to do is modify my server accordingly, and the client does not change.

Does this answer your question?

Joe Pluta
www.plutabrothers.com

> -----Original Message-----
> From: thomas@inorbit.com
>
> On Tue, 13 November 2001, "Joe Pluta" wrote:
>
> > ODBC, no matter what the application, goes against every
> possible tenet of
> > distributed processing.  It is slow, and it ties your host
> database to your
> > client code.  You cannot change even the names of your columns
> (much less
> > the physical layout and location of your data) without updating
> your client
> > code.  This is absolutely unacceptable.
>
> I'm clearly missing something here. How do you go about changing
> your physical layout, etc., without affecting client code when
> you DON'T use ODBC?