If we consider continuous operation for the algorithm, we can have at the
beginning of the bin a record for our flow. If the slice spans the entire bin,
it counts everything exactly and thus the variance of all estimator is 0. If
the slice ends in the current bin, we can divide the flow into two parts: one
covered by this older record and the rest. For the first part we have 0
variance for the byte and packet counts and for the second part we can apply
formulas 7 and 8, but instead of 
being
the number of packets of the flow in the bin, it should be only the number of
packets in this second part and the 
be the sizes of those packets. For the
flow count estimate, if the number of packets in the first record is 0,
the variance of the estimate is 0,
otherwise formula 9 applies. Thus having flow records active
at the beginning of the bin does not increase the variance of the packet, byte
and flow count estimates, on the contrary, it can reduce them significantly.