...declines. 
The interested reader will notice that the NASA and Rochester curves stop in 2 and 3 weeks respectively. The reason is that in order to conduct a meaningful prefetching experiment, we need at least two time intervals: one interval to find the Top-10, and a second interval, to give the Top-10 to clients and measure their hit ratio. Since we had only one month's traces for NASA, the highest time interval we could simulate was 2 weeks.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Evangelos Markatos
Fri Nov 1 16:38:26 EET 1996