[nsp] I need ideas on how to simulate a 200ms round-trip latency and bandwidth limitation.

[lee.e.rian at census.gov]

At 06/02/2004 05:13 AM, Brian Vowell <brian at digitalix.net> wrote:

> On the same topic...
>
> Anyone have much experience with what Cisco calls Long Fat Networks
> (LFN)?  Specifically, I'm interested in the impact (if any) of
> implementing the "ip tcp window-size" command on a 1721 router.

Using up memory on the router??  The "ip tcp window-size" command changes
the TCP/IP window size advertised by the router - which is only going to
make a difference for traffic sent to the router.

The reason I used to change the window size was to allow quick 'n dirty
thruput tests.   Enable tcp small servers & run ttcp against the discard
port on the router.  With the routers' default window size of something
like 2K more time was spend waiting for an ack then was spent transferring
data.  Increasing the window size on the router let me get an estimate of
what kind of thruput a user should get across that link.  But then our
security office started running Nessus scans on the whole network and some
of the routers complained about tracebacks or somesuch so I ended up
disabling tcp small servers on all routers.

> We have several VSAT links overseas, and I'm thinking of pushing the ISP
> on the US side to implement this feature on their routers, but I want to
> be sure that it'll have an impact on performance.

I suspect the only noticeable difference would be when using ftp or rcp to
download new IOS software to the router.  Changing the window size on the
router is not going to make any difference for traffic going through the
router.

>  The Cisco docs
> suggest setting the value to 750000, but don't go into much detail about
> how they arrive at that figure.

The standard suggestion is to set the TCP/IP window size to bandwidth *
round trip time.
For a T1 link where it takes a ping 40ms the suggested window size is
1.5Mb / 8 = 187,500 bytes/sec
187500 * 0.040 = 7500 byte window size

Take a look at
http://www-didc.lbl.gov/TCP-tuning/TCP-tuning.html

HTH
Lee