IP Checksum Offload

Hello,

I am trying to understand IP checksum offload. I got into this while trying
to track down a networking problem. I noticed many "IP checksum
errors" (using iptraf). I think that sent me down the wrong path because I
am starting to believe this is normal since iptraf looks at the packets
before the checksum is actually generated. (I got that idea from an
ethereal FAQ.)

But this leaves me with a question -- why do I not *always* get IP checksum
errors? I can run a network test for hours without one checksum error, but
if I crank up the network activity I start to see them.

Could it be that checksum offload is only invoked when the network load gets
to a certain rate? This might make sense since I have read that although
checksum offload gives a performance boost it does so at the expense of
less effective checking (since the checksum is generated later in the
process). Might the driver turn it on and off as load changes?

SOME FACTS:
I have a Dell Precision 360 with a gigabit ethernet on the motherboard
(Intel using the e1000 driver).

SuSE 9.3 with SMP kernel (for hyperthread support).

Here is output from "ethtool -k eth0"

# ethtool -k eth0
Offload parameters for eth0:
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp segmentation offload: on

Thanks for any info or leads!

Bob