[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: About TCP
>
> At 10:18 AM 11/21/97 PST, [email protected] wrote:
> >
> >There are two major variants of TCP in the simulator. The "regular one",
> >and "full tcp", where full tcp has (nearly) the full state machine
> >a regular TCP has. Full TCP can handle connection establishment and
> teardown;
> >it doesn't actually transfer data, although that would be a tiny change.
> >The "API" to TCP is not quite there yet (there is something rudimentary),
> >but it's on the list. Bug fixes and changes to full-tcp's connection
> >establishment/teardown are still in progress [it works ok when SYNs aren't
> >lost, but in certain such cases it doesn't behave exactly as a real TCP
> >would].
> >
>
> [email protected] wrote:
>
> I'm not sure where Kevin is with his full-tcp version, but we have a version
> that we created from an earlier version that was fairly buggy and didn't
> have an "API". (We gave this back to Kevin and Sally and I think Kevin
> is fixing this up to conform to ns-2 standards, but Kevin can correct me
> on this.)
I rolled in most of the changes Kathie supplied into the original version
and have been fixing other bugs as I find them. A recent addition (also
still under development) to nsv2 is an enhancement to the ErrorModel
stuff. The new mechanism allows for several ErrorModels to be in place
on a link and to choose between them based on flow-based information.
This setup makes the creation of packet-drop scenarios {for testing
full tcp's connection establishment/teardown} quite a bit easier.
So, the current status is that I am adding a number of packet-drop
cases to the full-tcp test suite in order to stress the robustness
of connection establishment/teardown. Once it looks to be in stable
shape, I will make an announcement.
- Kevin