[omniORB] CORBA benefits over EJB model

Zed Shaw zedshaw@killnine.net
Sat, 9 Jun 2001 00:07:50 +0000


See, I told you I'd get in trouble for saying that.  Try reading it again, 
but like this, "Doing CORBA in C++ requires "Guru" status...."  My assumption 
was that, since he was trying to do parts of the system in a language that 
was faster, we would use a compiled language, and therefore the only compiled 
language that is commonly used and supports CORBA is C++.   My assumptions 
were wrong, I guess.

Thanks for the clarification.

Zed

On Thursday 07 June 2001 14:06, W. Eliot Kimber wrote:
> Zed Shaw wrote:
> > 2)  (I'm going to get in trouble for this one :-).  C++ requires "Guru"
> > status just to code simple CORBA servers.  EJB requires "coding moron"
> > status even for advanced stuff.
>
> The use of CORBA does not require C++. We have written a sophisticated
> CORBA-distributed app entirely in Python. The only C code we've written
> for this project is patches to our 3rd-party dependencies (and small
> ones at that). I am anything but a CORBA guru and I can go from abstract
> object design to running servants and clients in a matter of an hour or
> two (e.g., tasks where I am writing classes that expose existing
> functionality in our underlying domain objects, so that most of the work
> is defining the IDL and writing the servants, not implementing business
> logic). We have no C++ gurus on our team.
>
> One of the problems I've run into in talking to people about using CORBA
> is the misconception that CORBA == C++, which it most certainly does
> not.
>
> We've found that, especially using the Python CORBA framework stuff we
> developed (and have posted on the OmniORB site), that developing CORBA
> apps in Python is about as easy as it could be.
>
> While Python is not necessarily appropriate for all distributed
> applications, we have found it has allowed us to develop the first
> iteration of a working distributed system very quickly. Now we can
> iteratively re-implement components in C++ or Java as we need to to get
> the performance and security characteristics we need. But we have found
> that for relatively small scales (1-200 concurrent users) that our
> Python-based system is entirely acceptable (and we haven't yet squeezed
> out all the performance we can out of our Python code nor do we
> currently have the testing resources to determine the upper limit of our
> current scalability). Our system now has client components implemented
> in Python, VB (using Python as the COM-to-CORBA bridge), and Java using
> several different ORBs. DataChannel is primarily a Java shop, so we
> haven't had a requirement to do any direct C++ development of clients or
> servers.
>
> Cheers,
>
> Eliot

-- 
Zed A. Shaw