Re: [ns] Trivial question about TCPSink

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

On Thu, 26 Apr 2001, Srihari Raghavan wrote:

> I am using a variation of the traditional dumbbell shaped topology with
> both UDP (CBR) sources and TCP (infinite FTP) sources. I am trying to mix
> these two traffic in a common bottleneck link that has less capacity than
> compared to the rate at which the sources generate traffic. When I increase
> UDP traffic I expect TCP througphut over time to decrease due to the TCP
> congestion control.
> 
> The question is that I need to collect bandwidth (in Mbps) at the
> TCPSink. With, Agent/LossMonitor sink for the UDP case, I use "bytes_" in
> the oft-repeated "record" function to collect statistics. With TCPSink I do
> not have a similar data member.

You need to hack TCPSink's recv() function in the tcp-sink.cc file to
record the data you want, after introducing your own bound variables.
For the actual bytes received, numbytes in the tcp-sink file should
probably be a bound variable - but isn't. (note that header overhead
isn't counted, and doesn't exist anyway.)

There's a pointer to a (very simple, because I'm lazy) hack on my ns
webpage to record the rather different application-level throughput by
adding a recvBytes() method. This should probably increment a bound
variable called numbytesdelivered_ or somesuch.

TCPSink is generally neglected IMO.

L.

> I tried using the TCP agent's "ndatabytes_", but it gives the sender's
> view of the TCP bandwidth. What I need is the reciever's
> (TCPSink's) view of TCP bandwidth recieved over time. Do I need to use
> "acks" at the sender? Or is there any other way?
> 
> Thank you for your time
> /Srihari
> 
> 

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