[PolyORB-users] Polyorb Performance

[Thomas Quinot]

* Sturm, Peter, 2010-04-13 :

> I'm testing the performance (data throughput) of the Polyorb version GPL
> 2009.

Interesting experiment!

> Therefore I modified the examples in examples/cobra/echo to send 2 MB
> string data multiple times to the server and transmit it back to the
> client.

It would be useful if you could contribute the sources of your benchmark
so that it can be reused to evaluate the performance impact of
configuration or code changes.

> The measured bandwidth was around 6 MB/sec (client & server on same pc
> system). This seems to be very poor compared with other CORBA
> implementations.
> 
> Testing with TAO results to a bandwidth of 200 MB/s.
> 
> Does anybody have experience of the Polyorb performance?
> Could it be a misconfiguration of the Polyorb installation?

Many factors can contribute to differences in performances. There are
some fundamental design choices in a generic middleware such as PolyORB
where some performance has to be traded off for functionality and
reusability (as opposed to optimization for a specific use case),
but that certainly wouldn't in itself justify a different of that
magnitude for bulk data exchange.

The tasking model and compile-time optimizations you are using could
make a difference. The specific data types used could also intervene, as
well as data handling in the client and server (as opposed to the
middleware). The choice of data types can also explain observed
differences, because some optimizations are very specific to certain
data types in PolyORB.

For example there is specific circuitry for sequence<octet> that is not
currently enabled for strings, and that could explain a visible
performance difference. Could you retry your experiment with a
sequence<octet> and let us know of the result?

Regards,
Thomas Quinot.

-- 
Thomas Quinot, Ph.D. ** quinot at adacore.com ** Senior Software Engineer
               AdaCore -- Paris, France -- New York, USA