Cisco → Testing methodology for routers using IPERF.

Evening all,
Having got my head around the IPERF basics, I'd like to test the routers I have in my lab, but I have no idea how I might go about this to give "real world" results.
I plan to test bare bones, then NAT/ACL and on through CBAC to VTI/VPN connections.
I have a bunch of 1841's so I'll use a PC hanging off each Fa port initially. Then I'll do the same with a switch and VLAN's in the middle, if there is any point... and I'll do router on a stick as well, again assuming there is any point.
I have a 2801 with CME and an 887 to test as well.
What should I do to generate "real world" traffic across the router(s) so that I can get some realistic results?
I'm assuming I can run this test across the internet as long as I can route / port forward the traffic appropriately. Is there any point in this? Would it give me a read on the ADSL speed?
I look forward to the answers and more questions!

what specifically do you consider "real world" results? are you referring to the types of traffic being pushed through the box or are you just referring to throughput numbers given a certain set of configurations?

q.

Other than having users sitting behind the gear and blasting netflix / p2p / web2.0 / etc traffic,
not sure how much more "real world" you can get. Or is your endgoal to just find out based on
the hardware / config where exactly performance tops out, assuming no other variables / limitations?

Regards

Both good questions.
I'm thinking that if I know the performance of the routers with just the config in the way then I have a baseline, and any users hanging off the router will get a share of what the router is through-putting. So based on that, I'm thinking I just need to config the routers as required and then run IPERF to see what I get. Any flaws in that as a basic "how fast is my router" question?

shouldn't be.
i'd run multiple tests of different lengths.
the other important factor will be the mix of packet sizes affecting the throughput. i'd look to break down your speed in terms of "packets per second". from there -- you can do the math on packet sizes and create a throughput formula.

q.

None that I can see either. If you REALLY want to give yourself a headache and/or cure a case of insomnia,
feel free to read up on the RFCs about traffic testing -- IMIX, RFC2544, RFC 6349, etc.

Regards

IPERF is not capable of doing reliable network performance testing due to the fact it can not control IFG (inter-frame gap). It is a TCP throughput tester, and you will always get widely varying results from different hardware running different operating systems at different latencies on the same exact network.

Real network performance testing requires a device that allows you to do tests LIKE RFC2544. That kind of testing requires low-level access to the hardware, and are generally a few thousand bucks depending on the feature sets and level of complexity desired.

I personally use the Spirent boxes (due to their relative simplicity compared with the competition, great support and wide feature set) anywhere from 1meg to 20gig from a small lunch-box sized appliance from a central console.

If you plan to argue with carriers and show that you are truly NOT getting the bandwidth guaranteed, you need to show that a high-fidelity appliance sending X packets per second at Y bytes per frame are experiencing Z drops per second indicating a bandwidth constraint / oversubscription / other NETWORK issue. Without some kind of quantified result it is almost impossible to pinpoint where the issue is.

Other companies use IXIA instead of Spirent