[omniORB] newbie quick queries

[FountainB at switch.aust.com]

Eric Damphousse wrote:
> In this scenario, I would have an object called createuser running
> on 100's of machines. I the manager would only have to select a
> subset of those machines from my local client, set you up, and I
> am done.

Terminology again - you don't have one object running on 100's of
machines, you have 100's of servants (objects), each running in a
process on a separate machine. All these servants implement the
same interface (class). Presumably they all register in a single
naming service, using different names.

The question here is whether you are happy to have the server
processes running constantly, whether or not they are in use. In
many cases this is perfectly acceptable, in which case you may
wish to run them as a daemon (unix) or a service (windows).
In some cases you may prefer to conserve system resources,
especially where you might have several different types of server
process and they are only used sporadically. In this case you might
use an implementation repository to launch the processes on
demand. You still have to have the impl repository running
constantly - something has to intercept the requests.

Implementation repositories are neat, and commercial ORBs
often make a big deal of them because they are not standardised
and present an area where ORBs can differentiate themselves
from each other. Having said that, I feel that they are a bit
overrated. I have never really felt that I was suffering from the
lack of an implementation repository. Possibly such a tool
would be really useful in some situations, but not in the problem
domains I have been involved in.

Bruce Fountain
Senior Software Engineer
Union Switch & Signal
Perth, Western Australia