Check out the new USENIX Web site. next up previous
Next: Flow count variance Up: Variances of estimators Previous: Packet count variance

Byte count variance

We can also compute the variance of the estimates for the number of bytes (we number the packet sizes $ b_i$ in reverse order with $ b_1$ being the size of the last packet and $ b_s$ that of the first one).

$\displaystyle VAR[\widehat{b}]=1/p\sum_{i=1}^s(1-p)^{i-s+1}b_i^2$ (8)

Note how this variance is strictly lower than the variance of results based on random packet sampling $ (1/p-1)\sum_{i=1}^s b_i^2$ (except for the case of a single packet flow). This shows that for byte counts too, flow slices are a better solution than ordinary sampling.



Ramana Rao Kompella 2005-08-12