[tg] IP fragmentation issue.

[Hugues VAN PETEGHEM]

Hi all,

I got an IP fragmentation problem using 'tg'. Here is the problem statement :

---------                                      ---------
| pc10 |--------------------------------| pc6 |
---------10.0.0.10               10.0.2.2--------

I'm trying to generate a simple data flow from pc10 (10.0.0.10) to pc6 
(10.0.2.2). Here are the 'tg' input files :

# pc10
on 0:15 tcp 10.0.2.2.80
at 5 setup
at 6 arrival 0.5 length 2000
packet 300

# pc6
on 0:15 tcp 0.0.0.0.80 server
at 1 wait

After a couple of minutes, I use 'dcat' to transform the log files in text 
files and here are the first lines of both files :

# pc10
(...)
Event Time      Type            Address         Id      Length
-----------------------------------------------------------------

5.010170        Setup
6.509204        Transmit        10.0.2.2.80     0       2000
7.009049        Transmit        10.0.2.2.80     2000    2000
7.508948        Transmit        10.0.2.2.80     4000    2000
8.008956        Transmit        10.0.2.2.80     6000    2000
(...)

>From the client side, everything goes perfectly. Each packet is sent to 
10.0.2.2:80 every 0.5 second. These packet have a 2000 bytes length.

# pc6
(...)
Event Time      Type            Address         Id      Length
-----------------------------------------------------------------

0.004769        Setup
19.742468       Accept          10.0.0.10.1207  Association 5
21.238312       Receive         10.0.0.10.1207  0       2000
21.737539       Receive         10.0.0.10.1207  2000    1448
21.737888       Receive         10.0.0.10.1207  3448    552
22.237457       Receive         10.0.0.10.1207  4000    1448
22.237830       Receive         10.0.0.10.1207  5448    552
(...)

>From the server side, the first packet is well received, but after that, 
everyhing is "wrong". I mean, it receive the data, but fragmented. 

I think it is a 'tg' (or 'dcat') problem. Have someone faced this before ? If 
not, could you try it and tell me if you have the same problem.

Thanking you in advance :

H
-- 
Hugues VAN PETEGHEM
Institut d'Informatique - FUNDP
---------------------------------------------
Web : http://www.info.fundp.ac.be/~hvp/