em forwarding performance (was Proposed 6.2 em RELEASE patch

[Mike Tancsa]

At 08:09 AM 11/22/2006, Jeremie Le Hen wrote:
>Hi Mike,
>
>Thank you for spending that much time for benchmarking, this is really
>interesting.

Hi,
         More to come, and if you can think of other tests let me 
know.  Next is VLAN performance.


>Though this is a little bit off topic, I'm quite puzzled by the fact
>that having filtering rules on Linux or not doesn't change the result
>much.  NetFitler keeps track of *all* connections even if there are no
>ruleset loaded -- you don't have to ask for it

Not sure, but I would unload iptables from the kernel when 
testing.  I will check again today as I want to go back and test the 
LINUX kernel in UP mode to see what difference it makes.


>It would be interesting to know the real performance of Linux as a mere
>router if we want a true comparision with FreeBSD performances.

As just a router, they seem fairly close without any firewall 
rules.  RELENG_4 seems to be the only clear leader in terms of raw 
pps, at least in these tests.  I am still puzzled by the fact that 
ipfw does relatively poorly compared to RELENG_4.

I also have some PCIe bge nics I will try and test. There are some 
patches that Bruce Evans posted and I would like to see how they perform.

         ---Mike