[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Fwd: IP packet overlow at sending node.]



Hi,

some time ago I had this problem (attached) for whixh I did not receive
an answer. It came to my mind again and I was wondering if somebody
could help me out on this.

regards, Sjoerd.




Hi,

I've observed something and would like to ask the ns developers if there
is a special reason for the way the following is implemented:

consider two nodes, connected by one link, TCP sender on one node, TCP
sink on the other, unlimited number of TCP packets transmitted from
sender to sink. With the default
link settings (1Mb, 10ms, droptail queue) , no IP packets get lost at
the sender due to (output) buffer overflow. This is what shouldn't be
happening in real life either. If you change the default queue size from
50 to 2, IP packets do get dropped.

- Isn't this wrong? I guess that in a real TCP implementation, TCP will
only offer more data to IP if there is bufferspace. No IP packet should
get lost at the sender, irrespective of the linkspeed or output buffer
size.
- wouldn't it be better to use a queue in TCP and a callback mechanism
(such as described in chapter 7 of the ns notes and documentation) to
ensure that TCP does not transmit above its rate (and IP packet get
lost)?

Regards, Sjoerd Janssen