Re: new Iperf FAQ
minor nit: for 5(d) the text talks about 3 connections and the example
creates 5 (-P 5).
in 5(h) you might suggest cron for unix systems.
in 8, tools appears as to ols (right after diagnostic)
I'd add some text to 10:
TCP and UDP behave differently. TCP has mechanisms to ensure the delivery
of all data and to control it's sending rate in an attempt to fairly share
the path with other data streams. In contrast UDP does nothing to ensure
that receiver successfully receives the sent data and does not limit it's
data rate in attempt to fairly share the path.
As a result of both it's rebustness and fairness mechanisms TCP is
sensitive to many factors that UDP is not. It is often useful when bad
TCP performance is seen to run a UDP test to see if you can achieve better
performance. With UDP is is possible to measure the loss for a given
session, which is generally not true with TCP (but see <a
href="http://www.web100.org">Web100</a>). TCP is sensitive to loss
especially for high bit rate and/or highly latency paths (often called
long, fat pipes).
Might add a question early on, say 4.5:
Why is my TCP performance so poor? (Or I have my computer hooked up to
Gigabit ethernet and I'm only seeing 45 Kb/s! Why?!?!?)
TCP requires some tuning to obtain optimal performance. On most operating
systems the default limits on tuning TCP sessions severly limit the
performance you are like to see. See the <a
href="http://www.psc.edu/networking/perf_tune.html">PSC Tuning Guide</a>
I know you answer this indirectly elsewhere, but people aren't very good at
reading between the lines, especially when it comes to FAQs.
Could add to 15: (this is the stuff I'm doing right now, and actually I was
running Iperfs twice an hour over Abilene which is part of what you would
see on the netflow site. I'm not running them anymore.)
The <a href="http://www.teragrid.org">TeraGrid</a> project uses Iperf to
monitor acheivable bandwidth between the sites on an hourly basis. The
TeraGrid project regularlly sees 900+ Mb/s flows between NCSA (Champaign,
IL) and SDSC (San Diego, CA) on Gigabit ethernet connected hosts.
<a href="http://www.ncsa.uiuc.edu/About/Alliance/">National Computational
Science Alliance</a> use Iperf to monitor performance to and from partner
There are a lot of other folks using Iperf too.
On Mon, Jun 16, 2003 at 02:41:53PM -0500, Damon Cook wrote:
> Hello again,
> I've finished putting together a new version of the Iperf FAQ. Before
> putting it up as the final version, I'd like to get the opinions of the
> experts out there. Naturally, some things have been added or removed
> from the old version and some formatting has been done. I'd like to
> incorporate any comments that you have when I finalize the FAQ next week.
> Please email me back if you have any suggestions. You can view the draft at
> Thanks a lot. I appreciate your time.
> -Damon Cook
Jon Dugan | Senior Network Engineer, NCSA Network Research
jdugan --at-- ncsa.uiuc.edu | 269 CAB, 605 E Springfield, Champaign, IL 61820
217-244-7715 | http://www.ncsa.uiuc.edu/~jdugan/