Finally, we present in Figure 6 the numbers of soft faults
incurred by all the studied policies. The first column presents the
soft faults under constant threshold 
and variable modem pool
size. The second column presents the numbers of soft faults under
constant modem pool size and variable threshold. All parameters are
the same as for the corresponding plots in Figures 4 and
5. As we explained earlier, the number of soft
faults is not an accurate indication of the performance of a
policy--soft faults represent correct predictions that lead to
successful disconnections. Nevertheless, we show the soft fault
measurements for completeness. The reader can refer to these plots to
confirm that the predictions made by CIRG and LRU were very often
accurate--the number of soft faults is usually many times higher than
the number of hard faults.
Figure: Plots of soft faults for all traces under variable
number of modems (first column) and variable threshold (second
column). The parameters not shown on the plots are the same as
those used in Figure 4 for the first column and
in Figure 5 for the second column. The LRU
and CIRG curves often coincide and cannot be distinguished.
