For simplicity, assume that there is only one request to each track. Multiple consecutive requests do not change the final algorithm in any way.
Every track in cache has a time stamp that is updated with the
current time whenever the track is placed at the MRU
position of either list. Let denote the temporal length,
that is, the time difference between the MRU and LRU
time stamps of SEQ.
Let and
denote the rates of sequential misses of one
stream when synchronous and synchronous+asynchronous prefetching,
respectively, are employed.
![]() |
Figure 3 displays the behavior of and
as the temporal length varies. As discussed in
Section II-B, under synchronous prefetching, sequential
misses or its rate is inversely proportional to the degree of
read-ahead. If, however, the read-ahead are discarded before they
are accessed, the effective degree of read-ahead decreases.
Whenever he effective degree of read-ahead is less than the actual
degree, it is proportional to the temporal length of the list.
Hence, we have that