Re: [ns] Changing the header size for one-way TCP

[John Heidemann <johnh@ISI.EDU>]

>> On Thu, 29 Nov 2001, Neundorf Alexander wrote:
>> 
>> > 	Cool, after some years researching using ns you finally found
>> > that it doesn't add header sizes....
>> 
>> Any simulation is by necessity an abstraction of details not
>> considered important. ns was begun to investigate tcp algorithms,
>> rather than engineering numbers; a computer science approach rather
>> than engineering performance optimisation.

Lloyd's comment is right on target...ns by necessity makes
abstractions (as do ALL simulators); representing header sizes 
came late to the simulator because the TCP algorithm details were much
more important to (most) ns users.

>	The problem is that there is no layer architecture in ns, and that the development is kind of closed. After hacking ns for 4 months now, I think writing the basic simulator is a matter of, let's say some weeks at most. 

Taking those three comments apart:

>	The problem is that there is no layer architecture in ns...

How do you mean?  There is a modular architecture for things like TCP,
classifiers, wireless propagation, etc.  Parts are explicitly layered
(see the wireless pictures in the manual, for example).  Other parts
such as TCP use OO tools to explore variant protocols.

>...and that the development is kind of closed.

Again, we try to maintain a reasonably open development process.  For
example, daily snapshots and read-only CVS access have been around for
a long time, and we do our best to watch the mailing list for patches
(between the spam and the compiler errors :-).

(It's certainly the case that not all patches get integrated... we
have a very limited number of hacker-hours.  But we do take patches.)

>After hacking ns for 4 months now, I think writing the basic simulator is a matter of, let's say some weeks at most. 
>The heavy part are the more advanced algorithms, e.g tcp, routing, advanced queueing and so on. But they could be ported.

I completely agree with you here.  It's not that hard to write a
simulator---4 weeks is plenty of time to send packets around.
The hard part is supporting all those protocols.

But rather than replacing the ns kernel and then porting the other
code, I'd encourage you to submit patches about what you don't like
with ns.

   -John Heidemann