A few months ago I switched from using an old Watchguard Firebox II (Pentium
MMX 200, 64 MB RAM, uses dc driver for ethernet) to a Nokia IP110 (AMD Geode
266, 64 MB RAM, uses fxp driver for ethernet). I figured the boxes would be
comparable, and I was looking forward to using the fanless Nokia. My setup
uses a LAN segment, DMZ segment (OPT1), and the WAN.

I finally got around to doing a real-world throughput test using M0n0wall
1.2 on both boxes, and I was kind of surprised. I did all tests between the
LAN and DMZ, where I'm not constrained by my internet bandwidth. On both
boxes, I enabled polling since it is available. The rulesets were exactly
the same, and the test systems in each network were the same for the test.

Basically, the Firebox left the Nokia in the dust. Using iperf, I got
between 74-88 mbit/sec between the networks. It also left the WAN responsive
in other runs of the test (i.e. the box wasn't crushed). The Nokia did FAR
worse, never exceeding 28 mbit/sec in either direction (LAN->OPT1 or
OPT1->LAN).

The $64 question is -- why? I figure it has to be one of two things:

1) CPU. Even though it has the faster clock speed, is the Geode that much of
a dog? This is definitely part of the problem -- the CPU load spiked during
the transfers on the Nokia. The Firebox II's P200 is just a standard socket7
processor.
2) Ethernet. The Nokia has "Intel 82559ER Embedded 10/100 Ethernet" (fxp)
ports, the Firebox has "Intel 21143 10/100BaseTX" (dc) ports. Is there
anything in the driver implementation that would make it so lousy?

I guess there's a 3rd possibility -- the Nokia I have has some kind of
problem, but I don't think that's the case.

I do enough transfers between networks that I'm probably going to switch
back to the Firebox and deal with the fan noise.

Joe