next up previous contents
Next: A Component Performance Model Up: Performance Evaluation Previous: FIFOs   Contents

Sockets

In order to make a direct speed comparison of the upcall rate between a lean architecture and a CORBA implementation, we have to provide an implementation that uses TCP/IP communication. Communication using sockets is characterised by much higher latency than message passing using FIFOs. In figure 41 we can see that the lightweight toolkit performs 10 times better than a complete Orb (1.7 versus 17 kHz). In addition, it shows behaviour close to raw sockets (97%) which have been measured with the TTCP [Sla84] benchmark. Again, most of the overhead goes to dispatching. The Orb also spends a small fraction of time in polling the socket as the sender cannot fully saturate the receiver all the time. Here there is still room for further improvement. However, due to the Internet Inter Orb Protocol (IIOP) and presentation layer overhead, a performance closer to channel speed is not likely to be reached soon with complete CORBA compliant ORBs.


\begin{Figure}
% latex2html id marker 1117
[H]
\begin{center}
\resizebox{15cm}...
...on of socket communication, CORBA
and the OO toolkit.} \end{center}\end{Figure}



Johannes Gutleber
1999-10-29