Re: UDP parameters


On Tue, 13 Sep 2005, Rosanna wrote:

> I am running Iperf version 1.7.0 in 2 PC´s with 100 Mbps cards,
> connected through a switch. Both got Windows XP. I would like to do
> a simple throughput (bandwidth) test using UDP traffic. However I
> don´t know what values to use for parameters like: bit rate, amount
> of data to send between testing nodes or during how much time should
> I send a specific amount of data.

UDP is a close to an "empty" protocol, its purpose is to provide an
very low level system access to sending/receiving IP datagrams. One of
the only few difference from a performance point of view is the amount
of socket buffering, that you can tune thanks to the misnamed "-w"
(window)  iperf option. You definitely have to give a try to "-w" in
your tests, this may have an big impact.

There is no simple rule to set these parameters, except for the
duration of the tests, where a couple of seconds is usually enough,
being several order of magnitude longer than most transient events in
your system. These parameters are here so you can study the behaviour
of your system under different stress conditions and settings.

TCP is a totally different story, since it is designed to be
self-adaptative.


> I have noticed that if I increase bit rate, for example, throughput
> increases and, in some cases, also does lost datagrams.

This is probably how many people use iperf's UDP. To find this limit.


> So which is the balance point of this
> parameters to get a correct throughput measurement?

I don't think there is any "correct" throughput measurement. If you
could tell what is the "right" bit rate to give, then what would you
measure?

You need more background on UDP and TCP before fully understanding
what iperf can do for you.



Other Mailing lists | Author Index | Date Index | Subject Index | Thread Index