Discussion

Most of our experiments had been restricted to operate at 100Mbit because that was the data rate that we could use to complete them. When flow control was enabled on a 470T, there appeared to be a limit that the switch would impose on the number of packets that it would handle per second resulting in an effective data rate of about 100 Mbit in most cases. When flow control was disabled, the Alteon Gigabit NIC failed to start as it appeared not to be able to communicate with the 470T.

The Alteon Gigabit NIC also appeared to impose a limit on the number of packets that it would handle per second, irrespective of packet size. We do not know whether or not this is a driver problem as the version of Linux that we used was quite old. We also had an Intel Gigabit NIC on hand but could not perform similar experiments with it because there was no driver for it in Redhat 6.2 (but driver was available on Mandrake 8.0, which is based on the 2.4 kernel).

The data rate of an attacker appeared to be quite low. This might be accounted for by the fact that we used an unmodified Redhat Linux. We had no intention of increasing the data rate since less than 10 attackers in the current setup can already saturate a link.

We had also performed one experiment (Experiment 40) using the polling driver from the Click project and the result can be found in the Appendix. We were unable to obtain any improvement in performance as reported in the literature but could not invest any addition resource to investigate the reasons as we had already spent a lot of time and resources on the experiments already.