We also introduce and analyze one hybrid static algorithm called Divergence Caching, and a set of dynamic algorithms called Sliding-Window algorithms.
We show that the dynamic divergence caching algorithm is competitive, whereas the static algorithm is not.
They also indicate that if the request probabilities are fixed, then the DDC algorithm comes within 15-45% of the optimal SDC algorithm, that is, the static algorithm with the optimal refresh rate for the given request probabilities.
The solid line at 1 represents the average cost of the optimal static algorithm for each case.
The main result of these experiments is that the performance of DDC(k) improves sharply as k increases to about 23, and for k [is greater than or equal to] 23 the cost of DDC(k) is only 10-15% greater than the cost of the best static algorithm.
We also compared the optimal static algorithm to the better (on the particular run) of the static algorithms with refresh rates 1 and infinity.
For k [is greater than or equal to] 23, the average cost of DDC(k) is roughly 70% of the cost of the best static algorithm.
If the ratio of writes to reads remained constant over time, while the individual [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] varied, then the dynamic divergence algorithms were only slightly better than the best static algorithm.