With flow slices we have the same problem as with NetFlow if we want to perform traffic analysis using time bins: for flow slices that span time bins, we can only guess how many of the flow's packets were in each bin, and this introduces errors in the results. This problem is even more pronounced when analysis is required in very small time bins to capture more precise traffic dynamics. We can extend flow slices to support binned measurement of traffic by keeping multiple sets of byte and packet counters, one set for each bin the slice passes through. By keeping separate counters for each bin, the binning error is eliminated entirely, at the cost of increasing the size of the flow records. Note that the reporting bandwidth costs of this solution are significantly smaller than those of the solution used by Adaptive NetFlow where an entire record is reported for each bin. The byte and packet counters are 8 bytes whereas a complete record is 48 bytes.
The number of counters per record has to be one larger than the number of bins
required to fit a slice because the flow slice can overlap only partially with
the first and last bin. The choice of the size of the measurement bin supported
is a compromise between resource consumption at the router and accuracy of
results. Reasonable choices can range anywhere from the slice length 
to 20
times smaller. For brevity, we do not explore this further in the paper, but
note that depending on the final goal, the flow slicing algorithm can be extended
with additional resources to obtain the desired accuracy.