next up previous contents index
Next: Training Database Description Up: Experimental Results and Evaluation Previous: Experimental Results and Evaluation   Contents   Index


Real Traces for Training

For all our experiments, we used real network traces representing the total traffic collected from the ingoing and outgoing flows at the ``ENST de Bretagne (ENSTB)'', at Rennes, France during 6 months (starting from October 1999 to March 2001). The traces are sampled and averaged every 5 minutes. We show in Table 9.1 a portion of this data. As previously mentioned, other proposals used artificially generated traces by means of mathematical models or simulations which are easier to predict. Others used special kinds of traffic (MPEG) which has periodicity by nature due to the frame patterns IBBPBBPI..., which is easy to fetch for the NN. All these proposals works quite well for these special traces, but fail when used to test real traffics as shown in Section  9.4.

Table 9.1: Some entries from the ENSTB trace.
Sample number Incoming traffic Outgoing traffic Total traffic
1 1580 1073 2653
2 2353 841 3194
3 3636 1523 5159
4 1600 1329 2929
5 303 965 1268
6 217 555 772
7 250 639 889
8 254 1188 1442
9 260 4434 4694
10 263 888 1151
... ... ... ...


next up previous contents index
Next: Training Database Description Up: Experimental Results and Evaluation Previous: Experimental Results and Evaluation   Contents   Index
Samir Mohamed 2003-01-08