Re: TCPSink shrinking window on sender..

[Kedarnath Poduri <kpoduri@BayNetworks.COM>]

vishnu,
	
>Hi,
>
>The simulated TCP packet in ns does not contain the recv adv. window.  In
>fact this is specified as a sender parameter.
>
>Any particular reason for that.. 

In the initial versions of ns tcp model wasn't an exact replica of actual
TCP. Rather you had a tcp source that generates the packets and a tcp sink
that absorbs these packets after generating pseudo-acks. One could say its
a pseudo TCP. Hence recv adv window variable did not even figure in that
code. In the present ns code you have full-tcp which is very similar to a
real-world implementation of TCP. However it still does not have a API. In
a real-world TCP the application that is performing a read operation calls
the tcp_output to send a window update which inturn is used by the source
to adjust the rate. However here (in the present ns-2) we still do not have
a user process and hence receiver does not advertise window size.

>would it not be interesting to observe
>the behaviour of the sender when the recv shrinks and expands the
>advertised window.
>

It would be. Go ahead :-))

kedar

ps: If it is of any help - we have an implementation of tcp-full.cc that
has an API. 
(see www-nrg.ee.lbl.gov/floyd/tcp_init_win.html)
However you still have to modify the existing tcp code to incorporate recv
window variable.