[omniORB] CORBA 2.3 standard and fault tolerance

[Matthias Klein]

Jani Niittukari wrote:

> "There is no standardized mechanism by which applications can convey their
> reliability requirements to the ORB." Quotation from Book: Enterprise
CORBA,
> Dirk Slama, Prentice Hall.
>
> With couple days of studying i have realized that there is 3 ways to
> implement Fault tolerant CORBA system.
>
> 1. I wait for the CORBA 3 compliant ORBs (Mmm, can't wait)
> 2. I pick one of the "expensive" commercial ORBs (Visibroker, Orbix, ...)

Hmm. We have been investigating fault tolerance with CORBA for
some time. Did we overlook something? I cannot see how to realize
a fault tolerat system with Visibroker or Orbix (well, don't know about
Orbix2000) alone.
There is an ORB called HardPack (by Lockheed, or so i think) wich claims
to be completely fault tolerant.

> 3. I have to implement my own replication mechanism, which i don't want to
> do.
>
> If you know any more solutions, i would be pleased to see them.
>
That depends on what you really need: Fault Tolerance? That would mean
making faults completely invisible to "clients" of your system. Or some sort
of
"continuose availability": sometimes, a client gets an error, but retry will
succeed.
Do you need to share object state between clients? Has object state to
be preserved across faults?
The more of these questions are answered "yes", the more will be left to
be done by you. (Shared state is the hardest part, i think)

Cheers
Matthias