Char said:
If you have an opportunity to try that exact hardware with Win 7
instead of XP, you might be surprised. I was getting similar
performance with my cheap ($9) Trendnet Gb NICs until I installed
Windows 7. The performance nearly doubled, going to 990+ Mbps on a
regular basis.
I was doing an A/B comparison within the same computer. At the other
end, was the Windows 7 laptop running a Broadcom GbE. And my current
motherboard has a Marvell GbE (good) and the TG-3269 I was testing
(bad). The Marvell did around 117MB/sec goodput (user data bytes),
while the TG-3269 did around 70MB/sec. And by testing with various
computers, it was sensitive to CPU performance level. And
using the Performance plugin and monitoring interrupts, the TG-3269
had way too many interrupts. There were about five interrupts per
packet processed, which I would have guessed, is impossible. But
that's what my results showed. A shower of interrupts. That's where
the CPU cycles are going. And I don't think the TG-3269 has any
interrupt mitigation feature either. (None was shown in Properties
for the card - otherwise I'd have ticked it by now).
I agree with you, on the contribution the OS makes. My experiments
here, show Win2K has around a ~40MB/sec limit on performance. Even
the Marvell chip on my current motherboard didn't change that, the
last time I checked. But OSes later than Win2K, may be able to
get closer to full link rate. And I have managed to get close
enough to 117MB/sec enough times here, to be satisfied with
my other network components. It's just that card that let
me down.
I didn't start out wanting to benchmark it. I was doing a Gentoo
distcc build, and there was a performance graph, and I'd spent the
money to make the connections between the machines GbE. And
when the TG-3269 was put in the box that only had a 100BT NIC,
I didn't get much of a boost. And that's when I got curious.
That card sits in an antistatic bag now. It's function in
life, is to prevent the antistatic bag from blowing away.
Paul