Question on Iperf
> >Date: Thu, 16 Jan 2003 17:04:59 +0100
> >From: "=?iso-8859-1?Q?gianlucaboffa --at-- libero.it?=" <gianlucaboffa --at-- libero.it>
> >To: "=?iso-8859-1?Q?iperf_users?=" <iperf-users --at-- dast.nlanr.net>
> >
> >I already wrote an email to you to ask something about iperf, now i have
> another question for you.
> >I would like to know how iperf establish how much data it can send into
> the network, because i want to justify the use of the software in my work
> of thesis to valuate the available bandwidth.
> >Iperf does av-bw= sum of data sent * time, but how does it calculate the
> first term of the product?
> >Please send me your reply, it will be very important for me!!
> >Many thanks
Well it depends on the transport you are using in your test (UDP or TCP).
With TCP it uses the TCP mechanisms supported by the operating system to
determine how much data can be sent. With UDP it simply sends at a
constant (or as close as we can come) rate which you specify (or defaults
to 1Mb). I believe you are confusing the bandwidth delay product
calculation with determining available bandwidth. The Bandwidth * RTT is
the bandwidth delay product and helps one determine the optimal settings
of TCP buffers (window size) with the -w option. So it does not calculate
either term but they are determined by the operator of Iperf. If they are
testing a 10Mbit link and the two test nodes have a 10ms delay (as
measured with ping or some other utility) then buffers should be in the
100K range. Iperf determines available bandwidth by consuming as much as
it can then reporting how much it was able to send. If you tell a UDP test
to send more than the available bandwidth then the server will report how
much it saw and be a reading of the maximum throughput, but since Iperf is
designed to test what a application could get over a link this is a little
unreasonable since there will be potentially a lot of lost packets.
Kevin