Performance

One must remember that performance was not an issue in this project. However, it is interesting to have an idea of how fast a connection is established.

As Java programs are byte-compiled, the first execution of a method is very slow. The second execution is much faster; figure  shows that the RSVP process spent 180ms making a reservation for the first time, and it only spent 10ms processing a Path and a Resv message afterwards.

The following sequence has been repeated 1,000 times in order to prove the robustness of the code, and evaluate the time taken to process RSVP messages: $\EuScript{A}$ sends a Path message to $\EuScript{X}$, $\EuScript{X}$ replies with a Resv message towards $\EuScript{A}$, and then $\EuScript{A}$ sends a Path Teardown message in order to remove the reservation. Therefore, three RSVP messages are processed, and a connection is first established and then torn down.

The results are shown in table . The tests have been made on different workstations, running different JIT^4.1-compilers. Sun's JDK was used as well a TYA^4.2 which is written by Albrecht Kleine.

Table: Time spent to process 1,000 sequences of Path, Resv & Path Teardown

Machine	Java JIT-compiler
Sun UltraSPARC 1 143Mhz 128Mo running Solaris	Sun's JDK 1.1.4
Intel Bi Pentium Pro 180Mhz 256Mo running Linux	Sun's JDK 1.1.3
Intel Pentium 133MHz 64Mo running Linux	Sun's JDK 1.1.5
Intel Pentium 133MHz 64Mo running Linux	TYA 0.7

The time spent to process a message is less than 10ms. This value is greater than the value announced in section . The 10ms announced was the time spent to process a Path message and a Resv message, without having to read them. This function is not optimised, and extracting the message from a UDP packet would probably be a bit faster. Moreover, 3,000 messages are received in a row; the Java Virtual Machine sometimes has to stop processing them in order to do garbage collection.