Check out the new USENIX Web site. next up previous
Next: Replacement vs. Fixed Timeouts Up: Performing Replacement in Modem Previous: Metrics

Experimental Results

We performed several experiments with different disconnection policies, which we outline below. For all the experiments shown here, we set threshold tex2html_wrap_inline490 to be equal to tex2html_wrap_inline568 . This means that a user can be disconnected only if idle for more than tex2html_wrap_inline490 seconds, and the disconnection will be a hard fault if the user's total idle time is not at least tex2html_wrap_inline584 seconds. The results obtained with tex2html_wrap_inline586 are fully representative of all results we have observed for different values of tex2html_wrap_inline490 and tex2html_wrap_inline568 (i.e., we observed no systematic deviation when we experimented with tex2html_wrap_inline592 ).

For the CIRG algorithm, we used three lists of past idle times (i.e., the value of the parameter k of Section 3.2 is 3) and the predicted idle time was the average of the predicted time using the three lists. We also performed experiments with one and five lists and the results were very similar.





Yannis Smaragdakis
Tue Apr 25 15:09:47 EDT 2000