Re: [ns] Problem with monitoring TCP throughput

[Lloyd Wood <l.wood@eim.surrey.ac.uk>]

On Mon, 6 Mar 2000, Chris Develder wrote:

> Being a novice NS user, I posted a question last week concerning monitoring TCP
> throughput.

goodput. (actual packet/segment throughput is same or more.)

You don't have to modify if (numToDeliver); you can just override
recvBytes() instead, which is slightly neater.

The following patch to 2.1b5 is _really_ crude (but very accurate -
there's no sampling interval error, and if all you want are the
results for an application-level view of goodput...) and just dumps
the output to stdout in a form suitable for drawing in xgraph or
postprocessing for cumulative addition via the awk script at end:


$ more tcp-sink.h.patch 
92a93,94
>       // added by Lloyd
>       void recvBytes(int bytes);

$ more tcp-sink.cc.patch
192a193,199
> // added by Lloyd
> void TcpSink::recvBytes(int bytes)
> {
>       double now = Scheduler::instance().clock();
>       printf("%f\t%d\n",now,bytes);
> }
>

(I should just be able to use NOW here, of course...)

$ more transfer.awk 
BEGIN {
    total = 0;
}
{ 
    time = $1;
    transfer = $2;

    total = total + transfer;

    ink = total / 1024;

    printf ("%f\t%f\n",time,ink);
}


A better general solution would be to have a default recvBytes with
bound variable in TcpSink in 2.1b7, so that capturing goodput is
commonplace and widely understood. The sink is widely neglected...

L.

bound variables - we've heard of them.

<L.Wood@surrey.ac.uk>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>