On 4/25/05, Jacky C.K Tsoi <cktsoi at gmail dot com> wrote:
> Hi All,
>  I'm going to do some throughput testing of m0n0wall on various CPUs (maybe
> P2-450, P3-450, P4-1.6, etc...) to see which hardware platform would meet
> our requirements (we're going to have a dedicated 100Mpbs broadband
> connection). And we'd like to have more than 100Mpbs thoughput in average.
>  Anyone can give me some advices on doing this kind of benchmark? What would
> be the best packet size? any good software recommendation? anything to
> notice?

Iperf gives a decent indication of maximum single TCP stream
throughput.  On some recent benchmarks I've done (results to be
published soon), I also used NetPIPE [1].  It tests throughput with a
wide range of message sizes.  The maximum throughput from NetPIPE
ended up being roughly the same as the Iperf results.

I tested with a P4 1.6 and a PIII 733 as the two end points, connected
to the tested m0n0wall platform via crossover cables.  The two test
boxes have Intel PRO/1000 (em) NIC's, both running FreeBSD 5.3. 
Performance with a crossover directly between the two boxes was over
400 Mbps at 1000BT.  A VIA 1 GHz [2] was the fastest box I tested, and
it came out at around 93 Mbps.

If you're looking to get "more than 100 Mbps" throughput, you'll have
to test with gig NIC's if you use this type of testing.  You could use
4 10/100 NIC's with 4 testing systems and kick off iperf/NetPIPE tests
simultaneously on both sets of systems.  That's not an extremely
scientific test, but it'd give you a decent idea at least.  Iperf
would probably work out better in that situation since it's a timed
test so you know they'll both run for a set amount of time and finish
at the same time.

1 - http://www.scl.ameslab.gov/netpipe/paper/full.html
2 - http://tinyurl.com/b8dyf

-Chris