I need to test the network performance of several different PCs. All of these PCs have at least 2 gigabit Ethernet ports and I have a traffic generator/analyer to do the testing.

I've decided to do my testing using RFC2544 with the PCs set up as bridges. My main question here really is what can I do to maxmize my performance. I'm trying to test the hardware specifically so I want to make sure I disable any service or "feature" that effects performance negatively. I'm on a private test network so security doesn't matter either.

For example, when I generate traffic I have seen that with Red Hat 9 and Fedora Core 1 frames without valid IP information gets bridged, which is desired. With RedHat ES 4 and Fedora Core 4 the IP checksum must be valid for the frame to be sent. This means something is actually looking at the packets which has a big negative effect on performance. I have no idea where this filtering is happening though. I have iptables and ip6tables disabled.

When I install the OSes I choose everything and this disable some of the things I know I don't need. I'll post some results related to specific kernels and services as I figure them out.


So far the best performing PC is an older single CPU 2GHz Xeon with FC1. Newer dual 2.8GHz Xeons with the same NICs aren't performing half as well (with or without the SMP kernel) with FC4. I'd perfer to figure out why before I regress to FC1 or update the other PC to FC4.