[omniORB] BOA vs POA Performance

[Duncan Grisby]

On Monday 11 December, "Craig A. Roloff" wrote:

> Actually, catior shows the server's IORs contain the machine's IP
> address, not the loopback interface.  Per your suggestion, we tried
> running the client on a separate machine.  No change.  We even tried
> using the loopback interface (via -ORBpoa_iiop_name_port), also
> without luck.

Even though the IORs contain the machine's IP address, the OS should
be sensible enough to use the loop-back, rather than bouncing off the
network interface. Anyway, it doesn't matter, since that clearly isn't
the problem.

> Unfortunately, trying a different OS is not an easy possibility.
> We're willing to invest some effort to track down the problem but
> we're at a loss on how to proceed.  Any ideas?

Have you looked at the GIOP traffic dumps which you get with
-ORBtraceLevel 30?  If not, I suggest you do that, and see if there is
anything different between the 2.8 server and 3.0 server.

I assume you have done a straightforward port from the 2.8 version to
the 3.0 version, so you're not using any of the more complex features
like servant locators etc. They will add more overhead, but still
nothing like as much as you are seeing.

One thought of something to try, if possible, is to run a version
which does the same number of calls, but sends as little data as
possible in each call. That will show whether the time is spent
marshalling the data, or in dispatching the calls.

Other than that, I don't think I can help without seeing the source
code. Can you come up with a minimal (or at least small) example which
demonstrates the problem?  You can send it just to me if you don't
want to send it to the list.

Cheers,

Duncan.

-- 
 -- Duncan Grisby  \  Research Engineer  --
  -- AT&T Laboratories Cambridge          --
   -- http://www.uk.research.att.com/~dpg1 --