I have a strange problem during the throughput calculation. I have a FTP connection using TCP and TCPSink agents. I'm using a record trick to schedule logging to file every 0.1 seconds and the TCPSink bytes_ variable to retrieve the number of bytes retrieved (it gets zeroed out after every log). This trick seems to be extensively used by other people, like presented here:
http://mailman.isi.edu/pipermail/ns-users/2005-October/052150.html
Anyways, In cases where I have packet drops and traffic jams I'm observing some strange behavior. Immediately after the throughput drops to zero because of the traffic jam and packet drops, it spikes well above the total link bandwidth. This should simply not be possible. For instance, In a 0.1 second period after the traffic drops on the 1Mbits/s link (with 25 ms delay set) I get 22880 bytes. When put inside the formula I get the throughput of 1.83Mbits/s (22880 / 0.1 * 8 / 1000000 = 1.8304Mbits/s), which is much more than set bandwidth of 1Mbits/s.
I'd appreciate any information regarding this behaviour, why this happens in the simulation and how to properly calculate and draw this graph (I've read somewhere that the data might first be smoothed by using a smoothing algorithm. Is it true and will it be correct?). Thank you in advance.
Andrej
No comments:
Post a Comment